Technical Scheme for Cutting Seedlings of Cyclocarya paliurus under Intelligent Control of Environmental Factors

Abstract

Cyclocarya paliurus is a species with high economic, horticultural, and medicinal value. C. paliurus grows faster than other plants, increasing the demand for propagation through leaf and stem cuttings to produce seedlings. However, this species requires pre-control of environmental factors such as high temperatures (25–30 °C), humidity (80–90%), and specific light (2000 to 3000 lux) intensity levels during the cutting and seedling production process. However, it is difficult to predict suitable environments for the growth of C. paliurus. This study requires the use of big data technology to parameterize the method of intelligent control of the environment used in the process of making stakes and creating seedlings. Our main results were that an improved convolutional neural network and short long-term memory (LSTM) in big data technology were used with a new method, multipath hole convolution (MPCNN), to predict environmental factors in production of seedlings. Also, the research results show that the MPCNN and LSTM methods can accurately predict the necessary temperature, humidity, and light conditions in the production process of C. paliurus seedlings. For the prediction of environmental characteristics related to this species, the light characteristics have a high error distribution, but the method described here was able to accurately control this variation, with an error of less than 2%.

1. Introduction

Cyclocarya paliurus (Batalin) Iljinsk (family Juglandaceae) is an endemic Chinese tree species that is distributed in mountainous regions of subtropical China. C. paliurus exhibits a wide distribution across various terrains, including mountains, valleys, and limestone mountains, ranging from approximately 420 to 2500 m above sea level in southern China [1]. This species needs a precise control of environmental factors like high temperatures (25–30 °C), humidity (80–90%), and specific levels of light (2000 to 3000 lux) intensity during the process of producing cuttings and seedlings. However, it can be difficult to produce suitable environments in which to study the propagation of cuttings and seedlings. It has good adaptability to monsoon climates, although it is slightly drought-tolerant [2]. What makes this plant even more intriguing is the historical use of its leaves as a health food [3]. This traditional application could be attributed to the plant’s remarkable abundance of biologically active compounds. The leaves of C. paliurus boast a high content of these compounds, which could potentially contribute to its perceived health benefits. The exploration of these bioactive constituents holds promise for further understanding the medicinal properties and potential applications of this remarkable plant species.

C. paliurus is a majestic deciduous tree that can reach heights of up to 30 m. Resembling the foliage of Pterocarya (or wingnuts), it boasts pinnate leaves measuring approximately 20 to 25 cm in length, comprising five to eleven leaflets. Interestingly, the terminal leaflet stands out prominently among them. These leaflets span from 5 to 14 cm long and 2 to 6 cm wide. When it comes to flowering, this tree produces catkins. Unlike Pterocarya, where catkins appear individually, Cyclocarya paliurus forms clusters of male (pollen) catkins. On the other hand, the female catkins can grow up to 25 to 30 cm long, and upon maturity, they bear numerous small, winged nuts. These nuts feature a unique circular wing, measuring 2.5 to 6 cm in diameter, encircling the nut itself. This distinctive wing structure sets it apart from Pterocarya, which typically has two wings positioned on the sides of the nut [4].

C. paliurus is a highly prized and versatile tree species. Its leaves have long been recognized in China for their medicinal properties [5,6,7] and are frequently used in traditional medicine or brewed as a medicinal tea. As a result of its diverse applications, C. paliurus has become an important cultural symbol in China and is widely cultivated both domestically and abroad [8]. Its unique, pinnate leaves, which resemble a hand with fingers, have drawn the attention of botanists for years. This species survived and was preserved following the quaternary glaciation, and it is mainly distributed in southern China [9,10]. There are few species similar to C. paliurus plants, so it is a somewhat precious species [11,12]. This species is a tree-like plant with a tall, strong, hard, and tough trunk. Its height can be as high as 10–30 m, and the growth rate is relatively fast. C. paliurus generally grows in humid forests at an altitude of about 2000 m. When it comes to propagation, C. paliurus seed germination can be difficult, but with proper techniques, it can be successful [13]. Nonetheless, once fully established, the tree exhibits remarkable drought tolerance, making it an ideal species for dry and arid climates as well [14]. In fact, research has shown that C. paliurus has developed specific mechanisms to cope with drought stress [15], including the ability to reduce water loss through its leaves and rapidly activate the synthesis of proline and alleviate the production of reactive oxygen species (ROS) [16,17]. This impressive ability to adapt to challenging environments makes C. paliurus an exciting species for both horticultural and ecological purposes. However, as seedlings, the species prefers dark and humid environments, as they occur in the forest core.

Figure 1 shows the specific structure of C. paliurus and its leaf shape. This species also has a relatively high economic and garden utilization. Additionally, C. paliurus contains some uncommon trace elements, such as iron, zinc, selenium, chromium, and germanium. Additionally, C. paliurus is rich in polysaccharides, flavonoids, and terpenoids, which act to lower blood sugar [18] and inhibit liver cancer [19], as well as serve as a valuable nutritious plant [20]. Its cultivation method comprises cutting, grafting, layering, division, and sowing. Taking cuttings to produce seedlings has a more successful survival rate [21,22].

The cultivation and growth of C. paliurus requires strict environmental controls, which limits its reproduction methods. The method of raising seedlings via cuttings provides C. paliurus with a higher survival rate. Most cultivation starts with leaf and stem cuttings [23,24]. Furthermore, grafting techniques can be used to enhance the growth and yield of C. paliurus [17]. By grafting the resulting plant onto different rootstocks, growers can produce stronger, more disease-resistant plants with improved fruit production. The value of the method of producing seedlings is that it needs to cut stems, roots, and leaves from a mother plant as the basis for cultivation. These parts of the stems or leaves will become independent new plants under suitable environmental conditions. The method of cutting seedlings can easily proliferate excellent genes of the plant itself with high quality [25,26]. Moreover, the method of producing seedlings from cuttings can make the plant growth speed very rapid. Compared with other seedling methods such as seeding, cuttings have better genetics and better survival rates [27]. Seedling production from cuttings must be provided with specific temperature, humidity, and photosynthetic active radiation (PAR) requirements. If the planting substrate is too wet, it can cause cuttings and seedlings to rot. C. paliurus requirements for temperature, humidity, and PAR also vary with time and with changes in the external environment. This species requires a strict control of the growth environment, and it also has an important relationship with changes over time [28], which ensures fast growth and high survival rate. This study intends to use big data technology [29] to produce C. paliurus cutting and seedling production methods with intelligent control technology for environmental factors [30,31]. To do so, big data technology can determine the most suitable environmental relationship for cutting and seedling cultivation [32,33]. Big data technology can also study and extract the object features of the research content by evaluating large amounts of data related to the research object. There are many types of big data technology. According to the spatial and temporal requirements of the research object for features, this data can be distributed using a convolutional neural network (CNN) or long short-term memory (LSTM) methods [34]. There are many variants of neural networks, which effectively solve the needs for special cases such as the number of datasets and training time [29,35]. CNN has great advantages in processing the features of spatial data, and it can reduce the parameters in the feature calculation process. LSTM has good performance in processing time-related features; it can memorize and filter information for different time periods [36].

In this study, based on CNN and LSTM methods, an intelligent algorithm of multi-path hole convolution (MPCNN) and LSTM methods was designed to draw the intelligent control scheme of environmental factors into the process of producing C. paliurus cuttings and seedlings. These two neural network algorithms will extract and predict the appropriate environmental conditions such as temperature, humidity, and PAR that this species needs. The main goal of this study is investigating environmental factors associated with C. paliurus through five different aspects. Section 1 introduces the research background related to C. paliurus and the background related to big data technology. The background and techniques related to the development of the cutting method are analyzed in Section 2. It establishes an intelligent environmental factor control scheme for C. paliurus used during the cutting and seedling production process using MPCNN and LSTM methods in Section 3. Section 4 analyzes and studies the performance of MPCNN and LSTM methods in predicting and extracting relevant environmental features of C. paliurus. Section 5 then summarizes the intelligent environmental factor control scheme. This study presents a technical scheme for the cutting seedlings of C. paliurus under the intelligent control of environmental factors. The novelty of our research lies in the integration of intelligent control systems, advanced technologies, and precise regulation of temperature, humidity, and light conditions. This innovative approach holds great promise for improving the efficiency and success of C. paliurus cutting seedling production, contributing to its sustainable cultivation and conservation.

2. Related Work

The practice of grafting in agriculture has been revolutionized by labor-efficient automation, which has paved the way for wider adoption of this technique. Among cucurbit growers, the root pruned one-cotyledon grafting method, combined with the use of grafting machines, is the most commonly employed approach. However, success in grafting is not solely dependent on the use of machines; the cutting angle, which determines the compatibility of the rootstock and scion, plays a vital role in the survival and growth of the grafted seedling [27]. For example, Sun et al. [37] explored the broken head rate and stubble break rate of potato seedlings by the automatic machine for cutting and raising potato seedlings. This study also analyzed the kinematics and mechanical properties of this automated machine, and it also used regression orthogonal experiments to study the running mechanism and key components of the potato seedling machine [38]. Xu et al. [39] described that it is difficult to adjust the cutting angle of the cutting and seedling structure, which results in contact between the surfaces of the cutting rootstock and the seedling, affecting the survival rate and quality of the seedling growth. In this study, the contact area between cutting rootstocks and seedlings was explored using a visual image method which can accurately provide feedback for the real-time situation to cutting personnel. The results show that this automated visual image recognition method can ensure the contact surface between the rootstock and the seedling. It can also improve the survival rate and quality of the seedling. Jiang et al. [40] also determined that the structure of the traditional seedling cutting method can easily lead to the damage of seedlings. The operation process of the cutting and seedling raising robot also affects the quality and survival rate of seedlings. The results of the study showed that the improved method of producing seedlings from cuttings could increase the survival rate of pumpkins to 98%. Fu et al. [41] also found that the current vegetable cutting method or single-bead cutting method is used to produce seedlings, which requires a lot of time and manual seedling grasping. This method of cutting technology has low efficiency rates. It designed a whole set of cutting and seedling raising methods to study the cutting and seedling raising process for melon seedlings. It was reported that this method of producing seedlings from cuttings has an extremely high efficiency rate, and the success rate of cuttings and seedlings achieved has this technology approaches between 92% and 100% for different forms of melon seedlings.

As already presented above, C. paliurus has difficulty producing cuttings that will result in new plants. Thus, the success of the grafting process is of great importance. In herbaceous plants, the success of the process is usually higher compared to that of woody plants [42]. For example, Sandhya et al. [43] described that the grafting process of Murraya koenigii, a wood-like plant, was 66% effective using a native M. koenigii rootstock, but loss through this process was reduced by 75% when the rootstock was wood apple. The improvement in the grafting process is recognized worldwide [44,45]. In this sense, Hansson et al. [44] tested two grafting techniques, grafting from—via activators regenerated by electron transfer atom transfer radical polymerization (ARGET ATRP)—and grafting to—by copper(I)-catalyzed azide–alkyne cycloaddition (CuAAC)—and concluded that the polymer content on the surfaces increased with an increasing graft length for the substrates modified via the grafting-from approach, confirming the possibility to tailor not only the length of the polymer grafts but also the polymeric content on the surface. The use of polymers associated with V-shaped cuttings has been widely used with success in several woody species [45] such as orange [45], lemon [46], grapevine [47], and others. Zhou, Li, Li, and Li [45] conducted a study in which they experimented with cellulose nanocrystals modified with an amphiphilic triblock copolymer, PDMAEMA-b-PGMA-b-PHFBA. The researchers found that this method allowed for the design of advanced CNC-based materials that could benefit from both the intrinsic properties of cellulose nanocrystals and the new features conferred by the multifunctional grafted chains [45]. Their study concluded that this technique could lead to the development of sophisticated materials with enhanced properties, such as improved mechanical strength and thermal stability. Overall, the findings of this research demonstrate the potential of using cellulose nanocrystals modified with amphiphilic triblock copolymers in the design of advanced materials.

3. Intelligent Control Scheme of Environmental Factors for C. paliurus Cutting and Breeding

3.1. The Significance of Intelligent Algorithm for Control of Environmental Factors for Cuttings and Raising Seedlings

MPCNN and LSTM technology can be used to predict and extract the relevant characteristics of three factors—PAR, temperature, and humidity—during the cutting and raising of C. paliurus. These three characteristics in the process of produce cuttings and raising seedlings have a close relationship to changes in space and time. Different time periods have different parameter requirements for the optimal combination of the three factors. Compared with controlling the conditions manually, the intelligent algorithm can more accurately predict and evaluate the temperature, humidity, and PAR characteristics in the process of producing cuttings and raising seedlings. This is an efficient and accurate method. The technology and algorithms can also monitor the temporal relationship during the process of producing C. paliurus cuttings.

3.2. Control of Environmental Factors Using MPCNN and LSTM Methods in C. paliurus Cuttings and Seedlings

The intelligent control scheme algorithm uses MPCNN and LSTM methods for the intelligent control of environmental factors in C. paliurus cutting and seedling raising. It mainly controls the three environmental characteristics, temperature, humidity, and PAR, required in the cutting and raising seedlings. It will classify the collected datasets related to C. paliurus and classify all climatic features. Then, it will use MPCNN and LSTM algorithms in big data technology to identify and map these three features. Figure 2 shows the process of applying MPCNN and LSTM in the intelligent control of the C. paliurus cutting and seedling cultivation environment. First, it will collect extensive data about C. paliurus and the basis for the intelligent control of these environmental factors. Then, these data will be divided into the three types required for the intelligent control of environmental factors. When the temperature, humidity, and PAR features are classified, it will use MPCNN and LSTM algorithms to identify and extract spatial and temporal correlations. Additionally, it controls the related environmental factors in the process of C. paliurus cutting and raising.

3.3. The Related Theory of MPCNN and LSTM Algorithm

The environmental factors monitored by the intelligent controlled system that are related to C. paliurus cuttings and seedlings uses MPCNN and LSTM methods. The MPCNN method is a variant of CNN, which is a multi-path CNN neural network structure. Compared with the single-path CNN structure, it can extract spatial features such as temperature, humidity, and PAR from different positions and depths. The CNN structure will include convolution operations, pooling layer operations, and computation of loss functions. MPCNN provides estimates similar to the CNN method; it just utilizes different paths. Figure 3 shows the structural and computational process of the MPCNN method. In this study, two neural network paths are used to extract the characteristics of the intelligent control process of environmental factors. This approach also reduces the corresponding number of parameters.

There is a great relationship between the PAR, temperature, and humidity characteristics of C. paliurus found in the process of creating cuttings and raising seedlings. During the day, the PAR is stronger, and the temperature is higher, but the PAR is weaker and the general temperature is lower at night. All of these different climatic features affect the growth quality and survival rate of C. paliurus in the process of cutting and raising seedlings. The intelligent control of environmental factors also needs to evaluate the temporal correlation of environmental data, which is intuitively important to monitor and control the living environment of C. paliurus. This study will use the LSTM algorithm to extract temporal correlations in environmental factors, which are often used to identify temporal features of research objects.

3.4. Theoretical Derivation of MPCNN and LSTM Algorithms

The LSTM algorithm is able to extract the temporal features of the research object due to the existence of a gate with different structures which are responsible for different tasks. Equation (1) illustrates the structure of the forget gate, which uses the aspects of weight distribution to determine the influence of historical information. It also selects important historical information to input to each layer.

$$f_t=\sigma (w_f·\left[h_{t-1}, P_t\right])+b_f$$

(1)

Equations (2) and (3) introduce the structure of the input gate of LSTM, which adds useful information to the cell state. It not only inputs the information of the current moment similarly to CNN, but also utilizes the comprehensive influence of the historical information and the time-dependent information, and then it inputs important temporal features.

$$i_t=\sigma \left({\omega }_i\right)·\left[h_{i-1},P_t\right]+b_i$$

(2)

$$\check{C_t}=\mathrm{t}\mathrm{a}\mathrm{n}\mathrm{h}\left({\omega }_c·\left[h_{t-1},P_t\right]+b_c\right)$$

(3)

The update gate is also an important component of LSTM which continuously adjusts the comparison created from information at different points in time, affecting the output data of the output gate, which includes the task of extracting useful information from the current cell state to be presented as an output. Equation (4) shows the detailed structure of the update gate.

$${\overrightarrow{C}}_t=f_t x {\overrightarrow{C}}_{t-1}+i_t x {\overrightarrow{C}}_t$$

(4)

The output gate determines the output data of the CNN layer, which contains a historical and current moment dataset. This is a fusion of two kinds of data; Equations (5) and (6) show the structure of the output gate.

$$O_t=\sigma \left({\omega }_o·\left[{\overrightarrow{h}}_{t-1},P_t\right]+b_o\right)$$

(5)

$${\overrightarrow{h}}_t=O_t x \mathrm{t}\mathrm{a}\mathrm{n}\mathrm{h}\left({\overrightarrow{C}}_t\right)$$

(6)

After collecting the dataset related to C. paliurus, it is relatively complex, and it does not have labels for different environmental factor characteristics, which then requires data classification based on distance. Equation (7) shows the calculation method of the dataset classification, which determines the classification of feature data based on distance.

$${dict}_{ed}=\sqrt{\sum _{k=1}^m{\left(x_{ik}-x_{jk}\right)}^2}$$

(7)

The interactive process of MPCNN and LSTM algorithms is the process of determining the optimal weight parameters, which also requires understanding the error between the predicted value and the real value. A “loss function” is a way to calculate the errors from temperature, humidity, and PAR characteristics, as shown in Equation (8).

$$E=\frac{1}{2} {\left(d_{out}-O_{real}\right)}^2=\frac{1}{2} \sum _{k=1}^t{\left(d_k-O_k\right)}^2$$

(8)

The determination of the optimal weights and bias must be calculated using the gradient descent method. The gradient descent method is based on the gradient, which is the value of the derivative. Equation (9) shows the derivative calculation method in MPCNN and LSTM algorithms.

$${\mathsf{\Delta }\omega }_{ji}=-\eta \frac{\theta E}{{\theta \omega }_{ji}}$$

(9)

Equation (10) shows the activation function ELU chosen in this study. The activation function is complex to nonlinearized relationships.

$$f\left(x\right)=\left\{\begin{array}{c}x, x \ge 0\\ \alpha \left(e^x-1\right), x<0\end{array}\right.$$

(10)

Convolution is a mathematical operation that involves combining two functions to produce a third function that expresses how one of the original functions modifies the other. In the context of signal processing and image processing, convolution is used to filter or enhance an input signal by convolving it with a filter or kernel function. Equation (11) shows a form of convolution calculation.

$${\delta }^{l-1}=conv2\left(rot180\left(W^l\right),{\delta }^l{,}^{\prime {full}^{\prime }}\right){\phi }^{\prime }\left(v^{l-1}\right)$$

(11)

4. Result Analysis and Discussion

This article uses the MPCNN and LSTM methods in big data technology [29] to analyze and investigate the intelligent control performance of environmental factors in the process of creating C. paliurus cuttings and seedlings. It selects the temperature, humidity, and PAR in the environmental factors as the research objects, and the deep learning method extracts the relevant features and relationships among these three environmental factors. Using only MPCNN and LSTM algorithms, datasets are a driving force and source of learning and for feature extraction. The dataset has the greatest impact on the deep learning algorithm, the subset of machine learning which is essentially a neural network composed of three or more layers. Compared with the impact of hyperparameters, it directly affects the weight distribution of the algorithm and its generalization ability. Therefore, before utilizing a deep learning algorithm, it requires precise processing and preparation of the dataset. Deep learning algorithms also require datasets to have more feature variations, to help avoid overfitting. In this study, relevant climate data from Yunnan Province, China, were selected as the source of the dataset for the environmental impact of C. paliurus cuttings.

MPCNN and LSTM methods used in this study involve three characteristics, temperature, humidity, and PAR, in the intelligent control of C. paliurus cutting and seedling raising environmental factors, which are important characteristics affecting the growth quality. If the MPCNN and LSTM algorithms can accurately measure and predict the changes in the three eigenvalues, it is beneficial to the survival rate of C. paliurus and will provide the best feature combination value. This study analyzes the global forecast effect using the mean forecast error of environmental factor variables. The prediction of three environmental variables during the cutting and raising of C. paliurus. In general, the prediction errors of temperature, humidity, and PAR can all meet the needs of cutting seedlings, and the prediction errors of these three are 1.72%, 1.81%, and 1.94%, respectively. The prediction errors of the three cuttings were all within 2%. The PAR is a difficult environmental variable to control, and it has the largest prediction error. However, it has also met the needs of C. paliurus cuttings for PAR conditions.

Temperature has a great influence on the process of C. paliurus cuttings and seedlings, where this species has higher requirements on temperature. The extreme temperature (lower and higher) is not conducive to the survival rate and growth quality of C. paliurus, which requires precise prediction and control of the temperature in the process of cutting and seedling raising. Figure 4 shows the scatterplot of predicted temperature characteristics during the cutting and raising. It can be seen that the errors of most temperature characteristics are below 2%, there are only three sets of data whose error values exceed 2% (Figure 4), and the largest prediction error is only 2.54%. These three points of the dataset only occupy 10% of the 30 datasets. This fully demonstrates that the MPCNN and LSTM methods can efficiently capture the temperature changes and temperature requirements in the process of C. paliurus cuttings and seedlings, which, in turn, can accurately control the temperature characteristic factors in the process of cuttings and seedlings in C. paliurus (Figure 5).

Humidity mainly refers to the water demand of the soil required in the process of C. paliurus cuttings and seedlings. The environmental variables of humidity can fully guarantee sufficient moisture in the cuttings and seedlings preparation. If the soil contains less water, this is not conducive to water and nutrient requirements in C. paliurus. However, if the moisture content in the soil is relatively high, this will cause some difficulties for the root respiration of C. paliurus seedlings as well as in other non-waterlogged species [48,49,50]. Also, extreme air humidity is an unfavorable condition for the cutting process, which requires precise humidity control. Figure 6 shows the linear correlation coefficient distribution of humidity-based environmental factors predicted by MPCNN and LSTM methods which reached 0.9853. This linear correlation coefficient over 0.95 can illustrate the accuracy of MPCNN and LSTM algorithms in intelligently controlling humidity environmental factors, and it can also provide researchers with reliable predictions (Figure 7).

5. Conclusions

Cyclocarya paliurus is a rare and special species generally cultivated via cuttings and seedlings, and has a relatively high survival rate and quality. However, this species has variable requirements for living environmental factors, and these factors must be precisely controlled, which can improve the survival rate in the process of cutting and raising seedlings. Light conditions which are too high will affect the photosynthetic machinery after creating photodamage to the photosystems in C. paliurus, a species non-acclimated to higher PAR. Humidity provides enough organic matter and water for C. paliurus. Too high a humidity is unfavorable for C. paliurus cuttings and seedlings. However, the control of these environmental factor variables is difficult to conduct manually.

In this study, a novel MPCNN-LSTM neural network was designed based on CNN and LSTM methods to achieve a precise control of the extraction and prediction of environmental factors in the process of C. paliurus cuttings and seedling cultivation. This study analyzed the temperature, humidity, and PAR conditions required by this species as a research object. Through this research, it can be found that the MPCNN-LSTM method can accurately and efficiently monitor and adapt the temperature, humidity, and PAR conditions during the cutting production and seedling growth of C. paliurus. The errors of the MPCNN-LSTM algorithm in predicting and extracting the temperature, humidity, and PAR characteristics of C. paliurus cuttings were 1.72%, 1.81%, and 1.94%, respectively. These error distributions can illustrate the reliability of the MPCNN-LSTM algorithm in the process of creating C. paliurus cuttings and raising seedlings.

With climate variations becoming more pronounced as a result of global warming, the adaptability of C. paliurus has become increasingly intriguing. In addition to its use in traditional medicine, this tree has drawn the attention of researchers and entrepreneurs for its ability to withstand fluctuating temperatures and contribute to temperature control. Its resilience in the face of climate change presents an opportunity for sustainable solutions. However, care must be taken with the grifting process to produce seedlings and cuttings to ensure the responsible cultivation and use of this remarkable plant.