Validation of an IoT System Using UHF RFID Technology for Goose Growth Monitoring

Abstract

Regular weight measurement is important in fattening geese to assess their health status. Failure to gain weight may indicate a potential illness. Standard weight gain analysis involves direct contact with the animal, which can cause stress to the animal, resulting in overall negative impacts on animal welfare. The focus of this study was to design a smart solution for monitoring weight changes in the breeding of farm animals. The proposed IoT system with a weighing device equipped with RFID technology for animal registration aimed to minimize the negative aspects associated with measuring in contact with humans. The proposed system aims to incorporate modern approaches in animal husbandry and use these obtained data for the potential development of husbandry approaches for different breeds of animals and enhanced managerial decision-making within husbandry. The system consisted of three main components: a data acquisition system, a weighing system with RFID, and an environmental monitoring system. In this study, the RFID system accuracy for detecting geese in the weighing system environment was assessed. The entire system evaluation yielded a sensitivity of 95.13%, specificity of 99.89%, accuracy of 99.78%, and precision of 95.01%. Regression analysis revealed a good correlation between observed feeding and RFID registrations with a determination coefficient of R² = 0.9813.

1. Introduction

The goal of precision agriculture in livestock farming is to meet the needs of farmed animals and, consequently, satisfy the requirements of breeders, suppliers, and consumers [1]. Monitoring the condition and growth conditions of animals [2] offers the possibility of discovering properties that would not be possible to ascertain through conventional breeding methods. Changes in animal behavior can be utilized to identify their health and welfare. An accurate monitoring system allows for gathering information about individual animals and the social behavior of individuals and the entire herd [3]. Systems focusing on online animal behavior monitoring are often aimed at large livestock, especially cows and pigs [4,5,6]. However, research has also been conducted on small animals such as poultry [7,8,9]. Most of these studies are centered around hens and broilers, which are often housed indoors.

It is common for animals to experience stress during the measurement of certain physiological and qualitative parameters. Stress factors include manual animal handling, weighing the animal, or measuring the animal’s temperature using manual probes. Automated monitoring of animal behavior using non-invasive measurement methods, i.e., non-invasive sensors, improves the welfare of livestock [10], including both large (e.g., dairy cows) and small livestock (e.g., geese). These sensors can be used to measure various parameters of individual animals, including feeding behavior parameters. Since feeding behavior is an important indicator of animal welfare and health, it is expedient for commercial farms to automate it [11].

Modern digital technologies enable intelligent automation of operations in poultry farming, leading to easier and cost-effective poultry production [8], ensuring suitable conditions for efficient poultry production [7], and providing appropriate conditions for efficient poultry production. Lahlouh et al. [7], in 2020, developed a system with 97.00% accuracy for controlling hygrothermal parameters such as temperature, relative humidity, and contaminating gases. Proper monitoring of environmental parameters such as temperature, humidity, ventilation, and the farming environment is essential to ensure optimal farming conditions. Additionally, controlling these parameters can lead to increased productivity and reduced energy consumption [12].

Various technologies have been developed for determining and monitoring the position and movement, including satellite systems [13], magnet-based systems, Inertial Navigation Systems (INS), and radio frequency-based systems [14]. However, many of these methods are not suitable for animal localization, especially for small animals housed indoors. For instance, due to the external walls of buildings, the Global Positioning System (GPS) cannot localize indoor objects [15], INS may be susceptible to errors requiring sophisticated filtering techniques such as the Kalman filter [16], and highly precise magnetic technology is sensitive to conductive and ferromagnetic materials at low frequencies [17].

The introduction of modern technical systems with the IoT has brought significant effects on crop cultivation and livestock farming [2]. The growing popularity of Internet of Things (IoT) technology in recent decades has led to the rapid development of systems based on radio frequencies. Among these systems suitable for indoor localization are frequency modulation technology [18], ZigBee [19], Wi-Fi [20], Bluetooth [21], radio frequency identification (RFID) [9], and LoRa [22].

RFID technology is a useful means for creating predictive models for the health and well-being of animals, as well as for comparing the impact of different housing systems on animal behavior [23]. For monitoring purposes, LF, HF, and UHF RFID technology can be used [23].

Using RFID technology, it is possible to monitor animal behavior during feeding [24]. Brown-Brandl et al. in 2018 and Maselyne et al. in 2016 [5,25] found that the HF RFID system developed in their study could be used to measure the feeding patterns of growing pigs in a commercial or simulated commercial setting. Given that automatic RFID measurements provide reliable information about actual feeding patterns, they demonstrated that RFID systems have the potential to be used for future research purposes in animal monitoring. Adrion et al. (2018) [11] proved that UHF RFID is a suitable RFID technology for monitoring the feeding behavior of growing and finishing pigs.

The performance of a system utilizing RFID technology can be verified through video recordings and image analysis. By comparing video recordings with RFID registrations, parameters indicating the system’s performance can be determined. These parameters include sensitivity, specificity, accuracy, and precision [5,11,26]. In the HF RFID system presented by Maselyne et al. (2016) [5], the use of values per minute led to a sensitivity of 99.3%, specificity of 96.1%, and accuracy of 77.6%. In Adrion et al. (2018) [11], the average UHF RFID sensitivity was 49.7%, specificity was 99.0%, and accuracy was 97.9%. The highest achieved average sensitivity was 79.7%.

Although UHF RFID technology has already been applied to poultry [26], considerable attention has not yet been paid to the breeding of geese, especially to monitoring their feeding behavior. The aim of this study was to propose an IoT system that utilized non-invasive methods for monitoring goose individual weight, which can be useful in predicting imbalances in feeding behavior as soon as possible. The cornerstone of this IoT system was the use of RFID technology. The main goal of this study was to validate the effectiveness of using UHF RFID technology within an IoT system, enabling the detection of weight changes in individual geese. Furthermore, environmental changes were monitored using a weather station to assess the impact of the environment on goose feeding behavior.

2. Materials and Methods

2.1. Animals and Field of Experiment

The research methodology was derived from the framework outlined in a study conducted by Krunt et al. (2022) [27], which centered on ducks as the primary subjects. Concurrently, our study was implemented with geese under identical environmental conditions. The measurements were carried out from June to August 2021, specifically in Prague Horoměřice at GPS coordinates 50.1147819° N, 14.2174647° E. The length of the day was approximately 16 h, and that of the night was 8 h at this location. Moreover, the average temperatures were as follows: 17.9 °C in June, 18.9 °C in July, and 20.3 °C in August. As part of the monitoring, we addressed the role of smaller animal breeding concerning their use in systems with multiple animals. Measurements commenced with goslings of domestic geese (Anser anser domesticus) at five weeks of age and concluded at 12 weeks of age. The measurements were conducted using a fully autonomous measuring and evaluation system, which monitored and alerted us to anomalies during the growth of individual animals. In addition to regular one-week health checkups for the animals, this study aimed to monitor the animals without direct interaction.

Feeding was provided ad libitum through a pelleted diet containing 20% CP and 11.2 MJ/kg ME. Throughout the monitoring period, the geese had unrestricted access to water provided in the form of a water pond, which was part of the housing, similar to that described in [27].

Measurements using the IoT system were conducted at two levels. The first and primary levels focused solely on monitoring changes in the individual weight gain of geese using RFID technology. At this level, observation of geese using a webcam is also possible. The second and secondary level monitored environmental changes and their impact on the feeding behavior of geese using a weather station and sensors to monitor environmental parameters.

The monitoring of geese encompassed the entire lodge, with lodge dimensions of 5 × 13 m. The block diagram illustrating the monitored area of goose movement is depicted in Figure 1. This image is complemented by a webcam view (Figure 2) of the outdoor goose enclosure (a) and the covered part of the enclosure (b). The locations of the webcams are outlined in a block diagram. The covered part of the enclosure is marked as the “Feeding area/electronic scale” in Figure 2a, and the camera location is denoted as “Cam 1.” The outdoor enclosure is indicated in Figure 2b as the area with the “Water pond and Weather Station,” and the camera location is marked as “Cam 2”.

2.2. Measurement System—Hardware Equipment

2.2.1. IoT System

The system was built on wireless data transmission using the Wi-Fi standard between individual main data collection nodes. The Wi-Fi signal was distributed across the area using an LTE router, which also served as a tool for remote control of the measurement system. The goose monitoring system was designed as a modular platform consisting of three main parts: a data acquisition system for data collection and evaluation, a feeding area, and an electronic scale for monitoring the weight change during the consumption of the feed ration on individual measurement days, and finally, the Environmental monitoring system, which was tasked with monitoring environmental parameters. The block diagram of the individual parts of the entire IoT system is shown in Figure 3. A picture of the distribution board as part of the data acquisition system is shown in Figure 4, the feeding area and electronic scale are shown in Figure 5, and a picture of the weather station as part of the Environmental monitoring system is shown in Figure 6.

2.2.2. Data Acquisition System

The data acquisition system consisted of a TP-Link Wi-Fi router and a distribution board. The system was powered using a 230 V power supply. Figure 4 shows the distribution board, which consisted of circuit breakers (1), a 12 V source (2), a computer (3), an RFID module (4), an amplifier (5), and a measurement card (6). The 12 V source (2) powered the amplifier (5). Data were measured using an Advantech USB 4716 (6) 16-bit measurement card with a sampling rate of 200 ksps. The measurement card was connected to the computer via USB (3). The measurement card was utilized to measure the voltage from the strain gauge system for weighing animals. An ambient light intensity sensor was also connected to the card, and a relay for restarting the RFID module was connected to the digital output. These data were processed in a minicomputer (3) with an Intel Core i5 8259U Coffee Lake 3.8 GHz processor, Intel Iris Plus Graphics 655, 8 GB of RAM, and a 256 GB SSD. The computer was connected to the internet via Wi-Fi. The software that recorded and processed data from individual devices ran on the computer. The activation of the created software occurred when the operating system started. The distribution board also contained an RFID module (4) with an RFID reader, which was connected to the system via Ethernet. The amplifier (5) was placed in a shielded box.

2.2.3. Feeding Area and Electronic Scale

The feeding area and electronic scale, which are shown in Figure 5, consisted of a feed reservoir with a feed inlet (1), an RFID antenna (2), a feed outlet from a feed reservoir (3), and a weighing system consisting of strain gauges (4) and a weighing plate (5). The weighing system allowed the weight of small animals up to 10 kg to be determined. The dimensions of the weighing chamber were b = 360 mm in width, h = 500 mm in height, and l = 420 mm in length. The dimensions are given under the same designation in Figure 5. The chamber of the weighing system was attached to the main frame using deformation members in the shape of beams set with strain gauges connected to the full bridge (4). A total of four deformation members were used, which were placed in the upper corners of the weigh chamber. Each strain gauge was designed for a maximum weight of 5 kg. The strain gauge sensors were connected in parallel to one amplifier, which allowed the elimination of incorrect weighing. The amplifier amplified the input signal and gave the strain gauge bridge power with a constant current source. The constant current source was selected at 7 mA with regard to the load of the current source and its maintenance at the setting value. The change in load was determined on the basis of the voltage drop according to the change in resistance of the strain gauges under load. The signal was then amplified and recorded using an Advantech USB 4716 measurement card. A photoresistor for measuring light intensity and a temperature and humidity sensor were placed in the upper part of the feeding area so that it was out of reach of the geese. The temperature and humidity of the environment were measured using the TH2E IP temperature and humidity sensor.

The weight gain analysis system was designed for the detection of individual animals using wireless UHF RFID technology. For detection, a 4-channel UHF RFID reader module, Chainway UR4, with an integrated Impinj R2000 circuit was utilized. This module allows communication via a serial line RS-232 or Ethernet RJ45 EPC C1 GEN2/ISO18000-6C. It can optimally read up to 700 RFID tags per second, featuring four channels and an antenna connection via an SMA connector. The antenna used was a 5 dBi UHF RFID antenna.

Moreover, an RFID tag made of ABS material was employed, enabling communication on three different frequency types: Low Frequency (LF) with a working frequency of 125 Hz, High Frequency (HF) with a working frequency of 13.56 MHz, and ultrahigh-frequency (UHF) with a working frequency range in Europe of 865–868 MHz and in the USA and Canada of 902–928 MHz. UHF with a fixed reader read range of 1 m was used for measurement. The RFID reading module was connected via Ethernet to the LTE router, to which the measuring PC was also connected. This approach allowed the creation of more extensive measuring systems by connecting several RFID readers.

Each goose was assigned an RFID tag with a unique serial number, which was then recorded in the created database. The system recorded the current weight during feeding along with the corresponding time and associated it with the individual RFID tag serial number. Additionally, other measured values, such as light intensity, could also be assigned to these numbers. One RFID channel was dedicated to measurement, while the other three channels remained unoccupied.

In Figure 7, a detail of the UHF RFID tag attached to the animal’s left leg is shown. The location of the tag, in the form of a ring, was chosen for the left leg, as the antenna was mounted on the left side, preventing possible signal shielding by the goose’s body. The feeding area and electronic scale were equipped with a feeding section, allowing each animal to enter the weighing chamber with the correct body orientation and be identified individually. The goose feed served as an incentive for regular visits to the weighing area. The antenna was configured to measure only on the weighing system, and it had a small transmission power of 5 dBi. The antenna was shielded from the other side by the cage structure of the feed box. Therefore, only those geese on the weighing system were detected.

2.2.4. Environment Monitoring System

The environmental monitoring system comprised a weather station and separate sensors. The weather station (Figure 6) included an anemometer (1), an anemoscope (2), a rain gauge (3), a distribution board with the electronics of the weather station (4), and a photovoltaic panel (5). Data from the weather station system were generated every two minutes, with the obtained values being assigned to the values during the RFID registrations. The mechanical anemometer (1) utilized a WH-SP-WS01 wind speed sensor to detect wind speed. The sensor’s revolutions were detected non-contactly using a magnetic reed contact. The anemoscope (2), made up of a WH-SP-WD wind direction sensor, was designed to detect wind direction. The anemoscope (0°) was directed north. The sensor contained eight magnetic switches connected to a resistor with varying resistance values, allowing for a total of 16 different rotation positions to be indicated, as the vane magnet could switch two switches simultaneously. The rain gauge MS-WH-SP-RG (3) employed the principle of self-emptying for measuring rainfall using a tilting rainwater collector. The distribution board (4) with the weather station electronics housed a battery, thermometer, barometer, regulator, and a microprocessor with a communication module. An ESP-WROOM microprocessor with a communication module was utilized, and a DS18B20 digital sensor served as the temperature sensor. The module also included sensors for air quality, temperature, pressure, and humidity: BMP280 pressure sensor, CCS811 eCO₂ sensor, and Si7021 module for temperature and humidity. The weather station was designed as an independent system and was powered using a photovoltaic panel (6) with a peak power output of 4.5 W, using a Samsung ICR18650-26J, 2600 mAh, Li-Ion battery. Additionally, the system integrated a flip module for measuring precipitation, a pulse system for wind speed measurement, and a resistance system for wind direction measurement.

The interior and exterior areas of the enclosure were additionally equipped with temperature, humidity, and photoresistor sensors. This information was crucial for spatial arrangements, considering that one section of the breeding area was shaded by a roof while the other was not.

A Reolink RLC-522-5MP IP web camera was used to monitor the enclosure. Recording the enclosure with animals using the camera enables analysis of their behavior and activities throughout the day, helping to tailor the breeding system accordingly. Not all activities can be deduced from feeding analysis alone. The camera, coupled with image processing, facilitates the analysis of the animals’ movement activities during the day or over a specific observation period. This includes tracking their movements during rest and feeding. The IoT camera also served to verify the geese feeding on the weighing system. While RFID registrations provide information about a goose’s presence near the antenna, they do not capture the animal’s behavior at the site. Therefore, the camera allows for analyzing geese behavior at the feeding place, particularly during RFID tag readings.

2.3. Measurement System—Software Equipment

The software utilized in the Intel Core i5 8259U mini computer was Windows 10. To handle data measurement and recording, a custom application named “Husa v. 1.2.0.0” was developed using Visual Studio 2019. For data post-processing, the open-source software Scilab 6.1.0 was employed. The software was designed to extract data from the RFID module, the IP device for measuring temperature and humidity, and the Advantech USB-4716 measuring card.

The RFID module could accommodate 4 antennas. The application allowed the selection of any connected antenna to sense the proximity of the RFID tag. It was capable of functioning in an autonomous mode. The principle was based on triggering an event when the presence of an RFID tag was detected within the antenna’s range. Upon reading the RFID tag, the information from the tag was cross-referenced with the record of the measured animals, each assigned a specific RFID tag number. This ensured that only intended tags meant for measurement were recorded. Consequently, measurement occurred only if information was successfully read from the RFID tag of the geese. Upon event activation, individual data such as voltage from the weighing device using the Advantech card (converted to weight), current temperature, humidity, and ambient light intensity were loaded. During weight measurement, five measurements were repeated, and average data were saved.

Upon launching the application, establishing communication with the RFID module via Ethernet was made possible. After connecting to the device, a request was sent for the hardware (HW) information of the RFID module, including details about the firmware version, hardware version, module number, temperature, and the set RFID frequency standard. This command acted as a watchdog in the program, triggering a device restart if no response was received. After connecting to the RFID module, the corresponding USB measuring card was selected, and the input channel number was configured. Subsequently, the number of channels of the RFID module could be chosen. As part of the measurement, one antenna was connected to the first channel. Once set up, measurement could be initiated when the RFID module sends a signal and awaits the reaction of the RFID tags. After the response, a comparison was made with the stored identification numbers. A record was created with the measured weight, ambient temperature, and ambient lighting if a match was found.

In addition to data measurement and processing, the application facilitated sending notifications, results, and anomalies regarding the condition of the animals during feeding. The system dispatched them every day at a set time, either via email or as an SMS message. Among the measured anomalies were, for instance, low attendance at the feeding place by animals or the mutual interaction of animals during feeding, leading to more than one RFID registration.

2.4. Measurement Algorithm

The flowchart illustrating the data measurement algorithm is presented in Figure 8. Upon the goose’s arrival at the designated measuring point with the assigned RFID tag, the RFID tag is read. A valid code is identified in the database if the RFID tag is successfully read. This is followed by recording the goose’s serial number, date, and time. Subsequently, the weight of the goose is recorded multiple times, and data from the external environment (temperature, pressure, precipitation, humidity, light intensity) are stored. All data are associated with a single time record and the animal’s presence. If the RFID tag is not initially read, the system will make repeated attempts until a valid code is found in the database.

2.5. Calculation of the Measurement Accuracy of the RFID System

With reference to [5], the performance measures of raw data from RFID registrations were determined. Based on the following formulas, the sensitivity, specificity, accuracy, and precision of the RFID system for detecting feeding geese were calculated:

$$\mathrm{S}\mathrm{e}\mathrm{n}\mathrm{s}\mathrm{i}\mathrm{t}\mathrm{i}\mathrm{v}\mathrm{i}\mathrm{t}\mathrm{y}=\frac{\mathrm{T}\mathrm{P}}{\mathrm{P}}$$

(1)

$$\mathrm{S}\mathrm{p}\mathrm{e}\mathrm{c}\mathrm{i}\mathrm{f}\mathrm{i}\mathrm{c}\mathrm{i}\mathrm{t}\mathrm{y}=\frac{\mathrm{T}\mathrm{N}}{\mathrm{N}}$$

(2)

$$\mathrm{A}\mathrm{c}\mathrm{c}\mathrm{u}\mathrm{r}\mathrm{a}\mathrm{c}\mathrm{y}=\frac{\mathrm{T}\mathrm{P}+\mathrm{T}\mathrm{N}}{\mathrm{P}+\mathrm{N}}$$

(3)

$$\mathrm{P}\mathrm{r}\mathrm{e}\mathrm{c}\mathrm{i}\mathrm{s}\mathrm{i}\mathrm{o}\mathrm{n}=\frac{\mathrm{T}\mathrm{P}}{\mathrm{T}\mathrm{P}+\mathrm{F}\mathrm{P}}$$

(4)

where,

• TP is the number of true positives, representing the instances when RFID registration was obtained and, at the same time, the video confirmed its correctness;
• TN is the number of true negatives, indicating instances when RFID registration was not obtained and, simultaneously, the video confirmed its correctness;
• P is the number of positives, signifying the times when the goose was standing on the scale;
• N is the number of negatives, representing the times when the goose was not standing on the scale;
• FP is the number of false positives, indicating instances when the goose was not present on the scale, but RFID registration occurred;
• FP is the number of false positives. The number of sampling points (s) when the goose was not present on the scale, but the RFID registration occurred.

3. Results

3.1. System Validation and Monitoring Results

The observation period was from 20/06/2021 to 11/08/2021, totaling 52 days. This period corresponded to the fifth to twelfth week of the goose’s life. Weight data linked to a specific code were transmitted during RFID registrations. In

Table 1. Results of weight gain for goose number 4 during monitoring and relevant data from the indicated occurrence, i.e., date, time, image, and RFID tag code.

Geese Number 4	Date: 13/07/2021 (8th Week)
RFID tag code: E2 80 11 60 60 00 02 07 86 ED 9B 1C	Time: 09:45:12

, the left column presents a graph evaluating the average weight values of goose number 4 per day. The X-axis displays the age of a goose in weeks, and the Y-axis indicates the weight value in kilograms. The graph clearly shows weight decreases during the rearing period. These declines were caused by medical checkups conducted once a week. This interaction was a stressful factor for the geese, resulting in reduced food intake and, consequently, weight loss. Weight loss, attributed to molting, was also observed during the eighth week of life. From the ninth week onwards, a decline in the weight gain slope is evident. The weight gain could be described by Equation (5), with a determination coefficient of R² = 0.9916.

$$y=-0.0428x^2+1.0329x-0.4953$$

(5)

The left column also displays the RFID code for goose number 4. In the right column, goose number 4 is presented in an image, marked at the moment it was on the weighing system, i.e., when it was detected by the RFID antenna. The right column also indicates the time and date of the given image.

From these collected data, the percentage of registrations occurring during monitored visits to the feeding system was calculated. Throughout the observation period, there were a total of 323,678 RFID registrations for geese that had just entered the weighing system, with an average of 15,804 ± 3809 (mean ± SD) registrations per goose. As the RFID tag of goose number 6 broke on 06/08/2021, goose number 6 was prematurely removed from the experiment. Therefore, it was also omitted from evaluating the average measurement value.

Table 2. The results from the weighing system for individual geese and the evaluation of the weight gain, overall number of registrations, and percentage of the time spent on the weighing system during the monitored period.

Geese Number	Initial Weight	Final Weight	Weight Gain	Number of Registrations	Percent **
1	3.45	7.40	3.96	15,348	1.02%
2	3.90	7.83	3.93	18,885	1.26%
3	4.23	8.01	3.78	19,170	1.28%
4	3.97	5.88	1.91	19,339	1.29%
5	4.40	7.55	3.15	18,462	1.24%
6	2.95	4.84 *	1.89	7598	0.49%
7	3.30	6.14	2.85	18,276	1.22%
8	3.87	8.35	4.48	15,740	1.06%
9	4.03	7.96	3.93	20,764	1.38%
10	3.68	7.49	3.82	11,709	0.78%
11	3.39	5.97	2.58	10,772	0.72%
12	3.56	7.95	4.39	19,980	1.33%
13	3.73	7.85	4.13	24,518	1.64%
14	3.97	6.69	2.71	11,372	0.75%
15	3.76	6.21	2.45	10,417	0.69%
16	3.50	6.95	3.46	13,406	0.90%
17	3.10	6.49	3.39	12,249	0.81%
18	3.70	7.02	3.32	13,028	0.87%
19	2.72	5.82	3.10	13,984	0.93%
20	3.31	7.42	4.11	15,027	1.00%
21	3.34	6.48	3.14	13,633	0.91%

* The RFID tag of goose number 6 broke on 06/08/2021. ** The percentage of time spent on the weighing system during the monitored period.

includes a comparison of the initial weight and final weight of individual geese and their weight gain during the monitored period. Subsequently, the table shows the number of RFID registrations for the observed period and the percentage of time from the total monitored period that the goose spent on the weighing system.

The number of RFID registrations made it possible to determine the time spent on the weighing system by multiplying the number of registrations by the length of the time step, which was 3 s. By multiplying these values for goose number six, it was found that the total time spent on the weighing system was 22,794 s. In contrast, goose number 13, which had the most RFID registrations, spent 73,554 s on the weighing system. Overall, goose number six spent the least time on the weighing system. The times spent on the weighing system were compared from 20/06/2021 to 04/08/2021.

A web camera was utilized to validate RFID registrations. With the assistance of the web camera, it was possible to ascertain whether the goose was actually present in the weighing system at a specific time. In evaluating the entire system, the sensitivity was 95.13%, specificity was 99.89%, accuracy was 99.78%, and precision was 95.01%. The indicated data were obtained from three monitored days (08/07/2021, 10/07/2021, and 13/07/2021) from 9:00:00 to 18:00:00. The resulting sensitivity, specificity, accuracy, and precision values for individual geese are presented in

Table 3. The results of validation parameters (sensitivity, specificity, accuracy, and precision) for individual geese.

Geese Number	Sensitivity	Specificity	Accuracy	Precision
1	99.31%	99.33%	99.34%	95.21%
2	93.68%	99.81%	99.62%	91.70%
3	95.00%	99.74%	99.41%	96.48%
4	95.10%	99.42%	99.13%	93.51%
5	86.38%	98.80%	98.44%	88.16%
6	100.00%	99.93%	99.93%	96.24%
7	100.00%	99.99%	99.99%	98.24%
8	96.15%	99.82%	99.68%	95.61%
9	96.47%	99.65%	99.35%	94.75%
10	99.39%	99.69%	99.67%	94.18%
11	98.96%	100.00%	99.98%	99.79%
12	95.91%	99.24%	99.01%	95.48%
13	89.69%	99.54%	99.13%	95.74%
14	96.09%	99.77%	99.58%	95.99%
15	99.79%	99.77%	99.78%	93.81%
16	94.81%	99.84%	99.46%	96.66%
17	97.55%	99.67%	99.60%	89.05%
18	93.70%	99.86%	99.75%	96.96%
19	99.36%	99.84%	99.84%	96.09%
20	97.89%	99.95%	99.92%	98.45%
21	93.39%	99.94%	99.16%	96.76%

.

Figure 9 and Figure 10 show a graph comparing the average time that geese spent on the weighing system each day. All twenty-one geese were compared over two chosen weeks. The first chosen week was in the period from 21/06/2021 to 27/06/2021, from the second day of observation. This was chosen because some geese started feeding on the second day, likely due to the stress caused by transportation and a new environment. The total time the geese spent on the weighing system on the first day of observation, 20/06/2021, was only 11,754 s. In comparison, they spent 22,704 s on the weighing system on the second day. Figure 9 shows a comparison of the average daily time spent on the weighing system during the first week. The X-axis shows the goose number, and the Y-axis shows the time the goose spent on the weighing system. The standard deviations for individual geese are also indicated in the graph. It is evident from the graph that goose number 6 spent the least time on the weighing system. This goose spent an average of 562 s with a standard deviation of 172 s, and its weight gain was 0.401 kg. Although her frequency on the scale was the lowest, her weight gain did not correspond. The lowest value of weight gain was for goose number 7, with a value of 0.234 kg; its average time spent on the scale was 942 s, and its standard deviation was 264 s. The highest weight gain was for goose number 2, which was 1.011 kg; this goose had an average time of 1102 s and a standard deviation of 278 s. Thus, the low value of RFID registrations for goose number 6 could be due to a faulty RFID tag or less need to spend time on the weighing system without the intention of feeding. The most time was spent on the weighing system by goose number 9, which spent an average of 1995 s. The weight gain of this goose was 0.399 kg, which was a relatively lower value compared with other geese. The goose could spend time on the scale even without her needing to feed. Goose number 9 shows a significant standard deviation value (σ = 1345 s). This was due to significant variations in the time spent on the weighing system during the week. The shortest time a goose spent on the weighing system was 612 s on 26/06/2021. The longest time a goose spent here was 4368 s on 27/06/2021. The smallest standard deviation was for goose number 16 (σ = 105 s). The shortest time spent was 573 s on 26/06/2021. The longest time a goose spent was 864 s on 21/06/2021. The average time goose number 16 spent on the weighing system was 693 s.

The second chosen week for comparing the average time geese spent on the weighing system was the period from 29/07/2021 to 04/08/2021, which was approximately the sixth week of monitoring. This period was chosen because the RFID tag of one of the geese (Goose number 6) broke on 05/08/2021. From Figure 10, it is evident that goose number 13 spent the most time on the weighing system, averaging 1337 s with a standard deviation of 347 s. The weight gain of goose number 13 was 0.465 kg. The lowest was 729 s, and its standard deviation was 484 s. The highest weight gain was for goose number 9, 0.801 kg; this goose had an average time of 1056 s and a standard deviation of 220 s. Goose number 21 spent the least time here. This goose spent an average of 513 s on the weighing system, and its weight gain was 0.677 kg. The graph shows that the lowest standard deviation was for goose number 7, with a value of σ = 1225 s. The average time spent on the weighing system was 1227 s. The shortest duration a goose spent here was 480 s on 01/08/2021. The longest time was 4194 s on 03/08/2021. The smallest standard deviation was for goose number 10 (σ = 61 s). The shortest time spent on the weighing system was 483 s on 04/08/2021. The longest time a goose spent on the weighing system was 660 s on 29/07/2021. The average time was 574 s.

In the week from 29/07/2021 to 04/08/2021, there was 19.99% less time spent on the weighing system compared with the week from 21/06/2021 to 27/06/2021. The difference in frequency was determined by comparing the total time spent by all geese in the first week and the total time spent in the week at the end of the monitoring period. The total time in the week from 21/06/2021 to 27/06/2021 was 154,668 s. The total time in the week from 29/07/2021 to 04/08/2021 was 123,744 s. This difference could be attributed to the fact that the geese no longer needed to be fed as they were at the beginning of the breeding period, corresponding to the observation that the weight of geese in this period was not increasing as rapidly as up to the ninth week of breeding (Table 1). Furthermore, it was an open breeding ground, allowing the geese to graze freely on available vegetation.

3.2. Comparison of Video Duration and RFID Registrations

Figure 11 shows a comparison of how RFID registrations (The black bar) correspond to actual goose presence on the weighing system (The red bar). Based on these comparisons, values for sensitivity, specificity, accuracy, and precision were calculated. The occurrences detected from the video recording are indicated in red in the graph, while the RFID registrations are indicated in black. For illustration, a time interval of 1 h from 10:30:00 to 11:30:00 on 13/07/2021 was chosen for geese 3, 5, 10, 9, 13, and 16. In the video evaluation, only instances where the goose stood with both feet on the weighing board were considered positive results. This caused falsely positive results in RFID registration, for example, when the goose stood on the weighing board with only one foot. This case occurred, for example, with goose number 10, where it is evident that the goose left the weighing chamber at 10:48:44 and returned at 10:48:57. Nevertheless, the RFID antenna continued to register her absence for 13 s. The opposite case occurred with goose number 5, who, according to the video recording, arrived at the weighing board at 10:37:27 and left at 10:38:07. According to RFID registrations, the goose arrived at the weighing board at 10:37:28 and left at 10:37:46. It is evident that the goose was on the weighing board; however, the RFID antenna did not detect it. This case could have occurred due to shading, manufacturing quality, or incorrect placement of the RFID tag.

Figure 12 provides a detailed comparison of RFID registrations (The black bar) and actual occurrences from the video recording (The red bar) of goose number 9. It is evident from the graph that, although the goose left the weighing system twice, the RFID antenna continued to detect its RFID tag. The RFID antenna detected the tag because the goose was standing at the edge of the weighing chamber. However, the recorded weight results using the software after reading the RFID tag did not correspond to reality. This incorrect result can subsequently be eliminated by removing weight values that are not realistic for the specific age of the goose.

Figure 13 shows linear regression for the dependence of RFID registrations on the time the geese spent on the weighing system. The time the geese spent on the weighing system is deduced from the video recording. The values were obtained from the monitored period from 9:00:00 to 17:00:00 on 13/07/2021. The total number of RFID registrations was 1110. The coefficient of determination for all geese was R² = 0.9813 with the regression equation of $\mathrm{y}=0.3272\mathrm{x}$.

3.3. The Weather Station

The Husa v. 1.2.0.0 software allowed for recording data upon detection of a goose’s RFID tag when it approached the weighing system.

Table 4. Data recorded and processed in Husa v. 1.2.0.0 software, obtained after RFID registration.

Weighing system
Date [dd.mm.yy]	Time [hh:mm:ss]	RFID code number		Average mass [kg]	Photoresistor [V]		Illumination intensity [lux]
09/08/2021	12:36:31	E2 80 11 60 60 00 02 07 86 ED E8 5D		7.279	0.481		1912.24
Weather station
Date [dd.mm.yy]	Time [hh:mm:ss]	Measurement ID	Precipitation [mm]	Wind velocity $\left[\frac{m}{s}\right]$	Direction [°]	Humidity [%]	Dew point [°C]	Temperature [°C]
09/08/2021	12:36:28	13375	0	4.16	270	57.8	14.5	22.5

presents the data collection results for a single RFID registration for goose number 1 on 09/08/2023 at 12:36:31. On 09/08/2023, goose number 1 approached the weighing system five times. One of the visits occurred from 12:36:10 to 12:37:04. The time 12:36:31 was selected from this visit for illustration purposes. The table is divided into two parts. The first part displays data obtained from the weighing system, which were saved as soon as the RFID tag of the goose was detected. The second part presents data from the weather station, saved every 5 s throughout the monitoring period.

From the table, it is evident that these data did not correspond temporally at that moment, as the time from the weighing system was recorded at 12:36:31, and the time gained from the weather station was at 12:36:28. The recorded time from the weighing system was within a five-second interval, during which the state of the surrounding environment was recorded, i.e., $12:36:28\le 12:36:31<12:36:33$. Each recorded data from the weighing system included the registration date, registration time, goose RFID code, the average weight value from 15 measurements taken every 0.2 s, and the voltage from the photoresistor. The photoresistor was connected to the circuit as a voltage divider. Therefore, these obtained data were in voltage values. The photoresistor was calibrated based on its wiring before being implemented into the experiment under laboratory conditions. After calibration, a calibration curve was determined, enabling light intensity calculation from the obtained voltage. The equation of the calibration curve was:

$$\mathrm{E}=295.81·{\mathrm{U}}^{-2.55}$$

(6)

From the table, it is clear that on 09/08/2021 at 12:36:31, goose number 1 was detected using the RFID code E2 80 11 60 60 00 02 07 86 ED E8 5D, with an average weight of 7.279 kg. At that moment, the voltage from the photoresistor, located in the sheltered part of the enclosure, was measured at 0.481 V, corresponding to a light intensity of 1912.24 lux. This RFID registration was associated with a record from the weather station on 09/08/2021 at 12:36:28, with a measurement ID of 13375. According to the record, there was no precipitation. In case of rain, data from the rain gauge would accumulate based on the number of tip-overs. The anemometer determined a wind speed of 4.16 m/s, and the anemoscope’s direction was 270°. The initial position of the anemoscope (0°) was directed north. Therefore, a direction of 270° corresponded to the anemoscope’s orientation toward the west. The air humidity was 57.8%, the dew point was 14.5 °C, and the temperature was 22.5 °C.

Figure 14a shows a histogram with the number of occurrences during the day and night, with night represented as a value of 0 and day as a value of 1. The time of day was calculated in the interval from 5:00 a.m. to 9:00 p.m. according to the current values of sunrise and sunset. The number of occurrences at nighttime (value 0 in the histogram) was 210 (45.36% of the occurrences), which corresponded to 8 h of the day. The number of occurrences during the daytime (value 1 in the histogram) was 253 (54.64% of the occurrences), which corresponded to 16 h. Figure 14b shows the histogram with the number of occurrences when there was wind and windlessness, with windlessness represented as a value of 0 and wind as a value of 1. The limit value of the wind speed was 1.5 $\mathrm{m}·{\mathrm{s}}^{-1}$, which was a value close to the median of all wind values from the evaluated period. The number of occurrences in the case of no during windlessness (value 0 in the histogram) was 288 (62.20% of the occurrences), and in the case of wind (value 1 in the histogram), it was 175 (37.80% of the occurrences). Data from the rain gauge showed that no occurrences of geese were registered at the time of monitoring.

The anemoscope data showed that the most frequent wind direction was southwest. In order to determine the relation between these data, a regression analysis (Figure 15) was performed, in which data from the anemoscope obtained during the goose registration (Frequency of wind direction in occurrences on X-axis) were compared with all data, i.e., also data where goose registrations were not recorded (wind direction frequency overall on Y-axis). As a result of the regression analysis, a coefficient of determination with a value of R² = 0.9792 was obtained.

Figure 16 shows the courses of the measured values of temperature (°C), humidity (%), and illuminance (lux) together with the cumulative distribution function for the occurrences of geese on the scales during individual days. In the graph, it is possible to observe individual occurrences both within the time of day and on the basis of the light intensity from the sensor located in the feeding environment. Light intensity ranges from 0 to approximately 3000 lux. The graph also shows the places (time intervals) on the distribution curve where frequent occurrences occurred, and these are the intervals when individuals visited the feeding place (the steeper the curve, the more intense the occurrences). On the other hand, places where the values do not change (time intervals without occurrences), can be observed on the distribution function. These data show that discharges are less frequent at lower night temperatures than at higher temperatures. It was also observed that during the morning and early morning hours, the geese spend time outside the feeding area (grazing on grass, etc.). Subsequently, during the day (higher temperatures), they spend time in the shade and at the same time visit the feeding place continuously until the evening hours.

4. Discussion

In this study, an IoT system was developed to monitor the growth characteristics of individual geese in a non-invasive way and, at the same time, control feed intake. The key element of the IoT system was the use of RFID technology, which is often aimed at large livestock, especially cows [28,29] and pigs [5,23,25,30]. However, research has already been conducted that focused on small livestock. RFID technology has been used to monitor chickens [9], which are given the greatest attention in poultry. In our study, the attendance of the domestic goose at the feeding site was monitored using UHF RFID technology. The designed system demonstrated high efficiency. The effectiveness of the proposed system was determined by parameters such as sensitivity, specificity, accuracy, and precision obtained by comparing RFID registrations with video recordings. Maselyne et al., in 2014 [31], using the HF RFID system, achieved a sensitivity of 88.58% and a specificity of 98.34% in pig breeding, and in 2016, a sensitivity of registrations of 99.3%, a specificity of 96.1% and an accuracy of 77.6%. Adrion et al. (2018) [11] obtained an average sensitivity of 49.7%, a specificity of 99.0%, and an accuracy of 97.9% using UHF RFID in pig breeding. When evaluating the entire system in our study, where, as in Adrion et al. (2018) [11], UHF RFID was used, the sensitivity was 95.13%, specificity was 99.89%, accuracy was 99.78%, and precision was 95.01%. These high-performance parameter values may have been achieved by choosing a good placement of the RFID antenna on the left side of the weighing system and positioning the RFID tag on the goose’s left leg, preventing signal shielding through appropriate tag and antenna placement.

The value added of the proposed system with RFID in our study was the placement of the RFID antenna on the weighing system at the feeding point, thereby determining the weight values that were assigned to the exact goose according to the serial number of the RFID tag. The feedstuff then served as an incentive for the geese to visit the area of the weighing system regularly. The advantage of this system was that, according to the radio frequency identification, the goose was easily recognized, and its weight was subsequently assigned to the recognized goose. This system made it possible to detect weight gains in individual geese without the necessary interaction with a human, which results in the elimination of an excessive stress factor, due to which food intake is reduced. From the graph in Table 1, regular slight decreases in weight during the monitored period are evident. These declines appeared after the animal’s health checks.

Our study analyzed the relationship between the feeding time, i.e., the time when the goose was registered using RFID, and the weight gain. As part of this analysis, two weeks of monitoring were compared, namely the first and approximately the sixth week. The results showed that in both compared weeks, the weight gains did not correspond to the time they spent at the feeding place. In the first week, goose number 2 had the highest weight gain value of 1.011 kg, while the time it spent on the weighing system was 1102. The range of average times for that week was from 562 s to 1995 s with an average value of 1052 s. Therefore, it is possible to say that the time the goose spent on the weighing system was rather average. The lowest value of weight gain was for goose number 7, with a value of 0.234 kg, and its average time spent on the scale was 942 s. Again, it was rather an average value of time. The same conclusions can be drawn from the second week of observation as well. At the same time, the non-relationship between the time spent on the weighing system with feeding and weight gain was supported by the results of the linear regression test for both weeks. In the first monitored week, the coefficient of determination R² = 0.0242, and in the sixth week, R² = 0.0001 was found. Spending a long time in the feeding and weighing area without significant weight gain could be an indication that the animal was resting on the weighing system. It is also necessary to consider the possible stress that prolongs the feeding process. According to Maselyne et al. (2016) [5], long and frequent visits could be an indicator of the animal being chased away and disturbed by other animals.

Intermittent weight changes in the measured animals were observed during data processing. These changes could be attributed to either a measurement system error or the movement of live animals during weighing. Variations in weight could also be caused by stress factors, the goose’s daily routine, and the feeding timing. After food intake, the goose’s weight might increase abruptly. Within this study, it was observed that the time spent on the weighing system with feeding in the seventh week of monitoring was 19.99% lower compared with the first week of monitoring, corresponding to the fifth week of the goose’s life. The developed IoT system allows farmers to monitor and provide information about feeding frequency in relation to a goose’s age.

Xue et al. (2023) [9] used UHF RFID to locate the exact position of chickens. The study was focused on monitoring within the framework of cage breeding, while the accuracy within a cage with dimensions of 40 × 40 cm was based on 88.74%. In our study, the proposed RFID system did not monitor the movement of geese within the entire housing but was only used to detect geese on the weighing system. A UHF RFID reader module, Chainway UR4, with four channels, was used for RFID detection. In this study, only one channel was utilized. After system expansion, the remaining three channels could be employed to monitor animal movement in specific areas of the rearing space and record passage. The entire system can be conceptualized based on triangulation measurement of animal movement. However, in the UHF RFID concept, it would be necessary to assess the suitability of use due to potential signal suppression by the animal’s body and the reason for the dimensions of the stables, which were 5 × 13 m, within the framework of open breeding.

A precise rearing system was employed in this study, providing information about the farm’s benefits and enabling the provision of comfort for the reared animals. These aspects lead to improved animal welfare with possible early recognition of potential animal illnesses. There are many studies with significant results that are situated in indoor or cage farms [9]. The created IoT system was able to monitor feeding behavior in animals and, at the same time, changes in the surrounding environment in open outdoor breeding.

The use of a local weather station is suitable because of the possibility of measuring local conditions relevant to the given location. The public weather station does not take into account differences in individual places. On the basis of local values, it is possible to analyze the behavior of the animals better and use it, for example, to adjust and adapt feeding processes. The results obtained from the rain gauge might possibly enable predictions of the need to replenish water and feed based on precipitation levels. For instance, if there were insufficient rainfall, potentially leading to drought and subsequent vegetation loss, the geese would be unable to graze. Consequently, the geese’s feed quantities would need to be increased. Unfortunately, from the point of view of precipitation, it was a relatively dry period within the monitored interval, so it was not possible to draw certain conclusions from our measurements, and this problem might be a subject for further studies.

In this study, the effect of wind on the feeding behavior of geese was also observed. Wind speed and wind direction did not show an effect on feeding behavior. Using data from the system, it was found that during the day, when the temperatures are higher, the geese spend more time in the shade and, at the same time, visit the feeding place continuously until the evening hours. Data from the meteostation, especially data on temperature and relative humidity, can be used primarily for the possibility of planning and managerial decision-making. Based on data and weather forecasts, the number of geese at the feeder and, for example, the cost of providing feed can be predicted.

5. Conclusions

In this study, an IoT system was developed that is capable of contactless monitoring of growth characteristics in individual geese. The created monitoring system allows for the observation of geese' feeding behavior while at the same time monitoring changes in the surrounding environment. The system enables the monitoring of weight within the flock without human intervention and, at the same time, checks the intake of feed. The change in feed intake is not visible on the geese until a few days later, but with the weighing system, the change in weight can be seen earlier. A system designed in this way can lead to breeding that allows an even increase in the animal’s weight without excessive weight fluctuations caused by stress factors and subsequently facilitate managerial decision-making for the length of breeding and the necessity to adjust the breeding process according to the needs of the geese obtained from the feeding behavior monitoring. When evaluating the UHF RFID system, the results showed a sensitivity of 95.13%, specificity of 99.89%, accuracy of 99.78%, and precision of 95.01%. Regression analysis revealed a good correlation between observed feeding and RFID registrations at the monitored time with a determination coefficient of R² = 0.9813. This study’s outcome was a smart farm aligned with the production and technical cycles of food production in a circular economy within Agriculture 4.0.