A method and system for predicting the growth stage of a crop
Patent Information
- Application Number
- CN202411104363.X
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-13
- Publication Date
- 2026-10-23
AI Technical Summary
[0002]中国作为全球最大的农业国之一，农业在国家经济和人民生活中扮演着重要的角色，在中国的农业国情下，人们对于粮食产量和质量需求越来越大，农作物的生育期是指从播种到收获的时间，它是衡量农作物生长发育的重要指标；农作物的生育期受气候、土壤、水分、肥料、病虫害等多种因素影响；农作物生育期的预测对农作物生产规划、病虫害防治、农业保险具有重要意义；然而，传统基于经验的生育期预测方法存在诸多局限性，主要依赖于农民长期积累的观察经验和历史气候数据，缺乏对复杂环境因素（如温度、湿度、光照、土壤条件等）动态变化的精确考量，导致预测结果往往具有较大的不确定性，此外，气候变化加剧了气象条件的不稳定性，使得以往的经验法则不再完全适用
[0018]本发明提供的技术方案中，通过安装在田间的传感器获取农作物不同生育期的生长环境参数数据，对所述生长环境参数数据进行数据标注及预处理，得到预处理后的生长环境参数数据；采用多模态CLIP的文本编码器对预处理后的生长环境参数数据进行特征编码，将经过预处理的生长环境参数数据转换成高维向量表示，得到文本特征；基于预处理后的生长环境参数数据获取对应的生长图像信息，构建图像编码网络，采用所述图像编码网络对所述生长图像信息进行特征提取，得到图像特征；通过温度线性融合模块，将所述文本特征和所述图像特征进行特征融合，形成一个综合特征表示；采用多层感知机预测模型预测农作物到不同生育期所需要的时间，通过构建最小均方误差损失进行模型训练，计算预测值与实际生育期时间差的平方和的平均值，评估模型的预测误差，并通过Adam优化算法调整模型参数；本发明通过预测农作物不同生育期的关键时间节点，能够及时筹备生产资料，采取更加科学规范的农艺措施，提升农作物产量。
Smart Images
Figure CN122965726A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of crop growth technology, specifically to a method and system for predicting crop growth period. Background Technology
[0002] As one of the world's largest agricultural countries, agriculture plays a vital role in China's national economy and people's lives. Given China's specific agricultural situation, the demand for both the quantity and quality of food is increasing. The crop growth period, referring to the time from sowing to harvest, is a crucial indicator of crop growth and development. The crop growth period is influenced by various factors, including climate, soil, water, fertilizer, and pests and diseases. Predicting the crop growth period is essential for crop production planning, pest and disease control, and agricultural insurance. However, traditional experience-based methods for predicting the growth period have many limitations. They rely primarily on farmers' long-term accumulated observational experience and historical climate data, lacking precise consideration of the dynamic changes in complex environmental factors (such as temperature, humidity, light, and soil conditions). This often leads to significant uncertainty in the prediction results. Furthermore, climate change has exacerbated the instability of meteorological conditions, rendering previous rules of thumb no longer entirely applicable. Summary of the Invention
[0003] The purpose of this invention is to solve the above-mentioned problems by designing a method and system for predicting the growth period of crops.
[0004] The first aspect of this invention provides a method for predicting the growth period of crops, the method comprising the following steps: Data on growth environment parameters of crops at different growth stages are acquired by sensors installed in the field. The growth environment parameter data is then labeled and preprocessed to obtain preprocessed growth environment parameter data. A multimodal CLIP text encoder is used to encode the preprocessed growth environment parameter data, converting the preprocessed growth environment parameter data into a high-dimensional vector representation to obtain text features; Based on the preprocessed growth environment parameter data, the corresponding growth image information is obtained, an image coding network is constructed, and the image coding network is used to extract features from the growth image information to obtain image features. The text features and image features are fused using a temperature linear fusion module to form a comprehensive feature representation. A multilayer perceptron prediction model is used to predict the time required for crops to reach different growth stages. The model is trained by constructing a minimum mean square error loss, the average of the sum of squares of the differences between the predicted value and the actual growth stage is calculated, the prediction error of the model is evaluated, and the model parameters are adjusted by the Adam optimization algorithm.
[0005] Optionally, in a first implementation of the first aspect of the present invention, the step of acquiring growth environment parameter data of crops at different growth stages through sensors installed in the field, and performing data annotation and preprocessing on the growth environment parameter data to obtain preprocessed growth environment parameter data includes: Data on growth environment parameters of crops at different growth stages are obtained, and the growth environment parameter data is labeled, missing values are handled, and standardized to obtain pre-processed growth environment parameter data. Set the neighborhood radius for each attribute in the initially processed growth environment parameter data, and calculate the neighborhood relationship matrix based on the neighborhood radius; For each attribute, calculate the neighborhood coarse entropy based on the corresponding neighborhood relationship matrix, analyze the importance of each data point in its neighborhood based on the results of the neighborhood coarse entropy, and calculate the importance of each data point and its relative neighborhood. For each data point, an outlier factor is calculated, which is obtained by weighted summation of neighborhood coarse entropy, importance, and neighborhood relative ratio. If the outlier factor of a data point is greater than the threshold, the data point is marked as an outlier and added to the outlier set. The outlier set is then removed from the growth environment parameter data to obtain the preprocessed growth environment parameter data.
[0006] Optionally, in a second implementation of the first aspect of the present invention, a multimodal CLIP text encoder is used to perform feature encoding on the preprocessed growth environment parameter data, converting the preprocessed growth environment parameter data into a high-dimensional vector representation to obtain text features, including: The pre-processed growth environment parameter data are arranged in order according to the time series and text labels are formed, where each set of data represents the environmental conditions at the corresponding time. The text encoder in the multimodal CLIP model is used to process the preprocessed growth environment parameter data through word embedding, position embedding and tag embedding. The word embedding, position embedding and tag embedding are added together to obtain the context representation. The context representation is converted into a feature vector of size 1×1024 using a self-attention mechanism. This yields text features.
[0007] Optionally, in a third implementation of the first aspect of the present invention, the step of obtaining corresponding growth image information based on preprocessed growth environment parameter data, constructing an image coding network, and using the image coding network to extract features from the growth image information to obtain image features includes: An image coding network is constructed, and growth image information obtained based on preprocessed growth environment parameter data is input into the image coding network. The first layer of convolution is performed using 3x3 kernels, 32 kernels, a stride of 2, and the ReLU activation function. Then, 64, 128, 256, and 512 kernels are used. Each layer uses 3x3 kernels, a stride of 2, and the ReLU activation function. With each additional layer, the depth of the feature map increases and the size decreases. The output feature map of the convolutional layer is subjected to average pooling through a pooling layer. When the feature map size is reduced to 16×16×512, a fully connected layer is then connected to obtain a feature vector of size 1×1024. Image features are obtained.
[0008] Optionally, in a fourth implementation of the first aspect of the present invention, the step of fusing the text features and the image features through a temperature linear fusion module to form a comprehensive feature representation includes: The temperature linear fusion module is constructed, and its calculation formula is as follows:
[0009]
[0010] in, This represents the recorded daily average temperature. , , These are the daily minimum temperature, daily maximum temperature, and daily average temperature, respectively, representing the three base point temperatures at different stages of crop development.
[0011] Optionally, in a fifth implementation of the first aspect of the present invention, a multilayer perceptron prediction model is used to predict the time required for crops to reach different growth stages, including: Construct a multilayer perceptron prediction model, wherein each dimension of the multilayer perceptron prediction model includes an input layer, a hidden layer, and an output layer. Let the input layer be represented by a vector. If this is the case, then the output of the hidden layer is f(W). 1 +b 1 The output of the output layer is softmax(W). 2 1 +b 2 ), where W 1 b represents the weight. 1 The variable represents the bias, and the function f represents the sigmoid or tanh function. The regression from the hidden layer to the output layer is softmax regression. 1 The output f(W) of the hidden layer is represented by 1 +b 1 ); The network structure formula of a multilayer perceptron is expressed as follows:
[0012] Where G represents the softmax activation function, and h represents the number of hidden layers. and This represents the weights and biases of the i-th hidden layer. This represents the input feature vector. and represents the weights and biases of the input layer, and S represents the activation function.
[0013] Optionally, in a sixth implementation of the first aspect of the present invention, the step of training the model by constructing a minimum mean square error loss, calculating the average of the sum of squares of the differences between the predicted value and the actual reproductive period, evaluating the prediction error of the model, and adjusting the model parameters by using the Adam optimization algorithm includes: The loss function uses the minimum mean squared error loss, and its calculation formula is as follows:
[0014] Where m represents the number of training samples. This indicates the time from the model's predicted growth period to the corresponding growth period in the crop's textual description, where T represents the actual number of days from the crop's growth period.
[0015] A second aspect of the present invention provides a crop growth period prediction system, which includes a data annotation module, a text feature encoding module, an image feature extraction module, a feature fusion module, and a model training module, wherein... The data annotation module is used to acquire growth environment parameter data of crops at different growth stages through sensors installed in the field, and to perform data annotation and preprocessing on the growth environment parameter data to obtain preprocessed growth environment parameter data. The text feature encoding module is used to encode the preprocessed growth environment parameter data using a multimodal CLIP text encoder, converting the preprocessed growth environment parameter data into a high-dimensional vector representation to obtain text features; The image feature extraction module is used to obtain corresponding growth image information based on preprocessed growth environment parameter data, construct an image coding network, and use the image coding network to extract features from the growth image information to obtain image features. The feature fusion module is used to fuse the text features and the image features through the temperature linear fusion module to form a comprehensive feature representation; The model training module is used to predict the time required for crops to reach different growth stages using a multilayer perceptron prediction model. The model is trained by constructing a minimum mean squared error loss, calculating the average of the sum of squares of the differences between the predicted values and the actual growth stages, evaluating the prediction error of the model, and adjusting the model parameters using the Adam optimization algorithm.
[0016] Optionally, in a first implementation of the second aspect of the present invention, the data annotation module includes a preliminary processing submodule, a first calculation submodule, a second calculation submodule, a third calculation submodule, and a data point marking submodule, wherein, The preliminary processing submodule is used to acquire growth environment parameter data of crops at different growth stages, and to perform data annotation, missing value handling and standardization on the growth environment parameter data to obtain the preliminary processed growth environment parameter data. The first calculation submodule is used to set the neighborhood radius of each attribute in the pre-processed growth environment parameter data, and calculate the neighborhood relationship matrix based on the neighborhood radius; The second calculation submodule is used to calculate the neighborhood rough entropy for each attribute based on the corresponding neighborhood relationship matrix, analyze the importance of each data point in its neighborhood based on the results of the neighborhood rough entropy, and calculate the importance of each data point and its relative neighborhood. The third calculation submodule is used to calculate an outlier factor for each data point, wherein the outlier factor is obtained by weighted summation of neighborhood rough entropy, importance and neighborhood relative ratio; The data point marking submodule is used to mark a data point as an outlier when its outlier factor is greater than a threshold, and add it to the outlier set. The outlier set is then removed from the growth environment parameter data to obtain the preprocessed growth environment parameter data.
[0017] Optionally, in a second implementation of the second aspect of the present invention, the image feature extraction module includes a construction submodule, a convolution submodule, and an average pooling submodule, wherein, A submodule is constructed to build an image coding network, which inputs growth image information obtained based on preprocessed growth environment parameter data into the image coding network. The convolutional submodule uses 32 3x3 convolutional kernels with a stride of 2 and ReLU activation function for the first layer of convolution. Then, it uses 64, 128, 256, and 512 convolutional kernels. Each layer uses 3x3 convolutional kernels with a stride of 2 and ReLU activation function. With each additional layer, the depth of the feature map increases and the size decreases. The average pooling submodule performs average pooling on the output feature maps of the convolutional layers. When the feature map size is reduced to 16×16×512, it is then connected to a fully connected layer, finally obtaining a feature vector of size 1×1024. Image features are obtained.
[0018] The technical solution provided by this invention acquires growth environment parameter data of crops at different growth stages through sensors installed in the field. This data is then labeled and preprocessed to obtain preprocessed growth environment parameter data. A multimodal CLIP text encoder is used to encode the preprocessed growth environment parameter data, converting it into a high-dimensional vector representation to obtain text features. Based on the preprocessed growth environment parameter data, corresponding growth image information is obtained, and an image encoding network is constructed. This image encoding network is used to extract features from the growth image information to obtain image features. A temperature linear fusion module is used to fuse the text features and image features to form a comprehensive feature representation. A multilayer perceptron prediction model is used to predict the time required for crops to reach different growth stages. The model is trained by constructing a minimum mean square error loss, calculating the average of the sum of squares of the differences between the predicted value and the actual growth stage, evaluating the model's prediction error, and adjusting the model parameters using the Adam optimization algorithm. This invention, by predicting key time nodes at different growth stages of crops, enables timely preparation of production materials, the adoption of more scientific and standardized agronomic measures, and the improvement of crop yield. Attached Figure Description
[0019] Various other advantages and benefits will become apparent to those skilled in the art upon reading the following detailed description of preferred embodiments. The accompanying drawings are for illustrative purposes only and are not intended to limit the invention.
[0020] Figure 1 A schematic diagram illustrating an embodiment of the crop growth period prediction method provided in this invention; Figure 2 This is a schematic diagram of the structure of the crop growth period prediction system provided in an embodiment of the present invention. Detailed Implementation
[0021] The terms “first,” “second,” “third,” “fourth,” etc. (if present) in the specification, claims, and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a particular order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms “comprising” or “having,” and any variations thereof, are intended to cover a non-exclusive inclusion; for example, a process, method, apparatus, product, or device that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or devices.
[0022] For ease of understanding, the specific process of the embodiments of the present invention is described below. Please refer to [link / reference]. Figure 1 A schematic diagram of an embodiment of the crop growth period prediction method provided by this invention. The method specifically includes the following steps: Step 101: Acquire growth environment parameter data of crops at different growth stages through sensors installed in the field, label and preprocess the growth environment parameter data to obtain preprocessed growth environment parameter data; In this embodiment, growth environment parameter data of crops at different growth stages are acquired. The data is then labeled, missing values are handled, and the data is standardized to obtain pre-processed growth environment parameter data. A neighborhood radius is set for each attribute in the pre-processed growth environment parameter data, and a neighborhood relation matrix is calculated based on the neighborhood radius. For each attribute, a neighborhood rough entropy is calculated based on the corresponding neighborhood relation matrix. The importance of each data point in its neighborhood is analyzed based on the results of the neighborhood rough entropy, and the importance and relative neighborhood ratio of each data point are calculated. An outlier factor is calculated for each data point, where the outlier factor is obtained by weighted summation based on the neighborhood rough entropy, importance, and relative neighborhood ratio. When the outlier factor of a data point is greater than a threshold, the data point is marked as an outlier and added to the outlier set. The outlier set is then removed from the growth environment parameter data to obtain the pre-processed growth environment parameter data.
[0023] Step 102: Use a multimodal CLIP text encoder to encode the preprocessed growth environment parameter data, and convert the preprocessed growth environment parameter data into a high-dimensional vector representation to obtain text features; In this embodiment, the preprocessed growth environment parameter data is arranged in an ordered time series and formed into text labels, where each set of data represents the environmental condition at the corresponding time. The text encoder in the multimodal CLIP model processes the preprocessed growth environment parameter data through word embedding, position embedding, and label embedding. The word embedding, position embedding, and label embedding are added together to obtain the context representation. The context representation is then converted into a 1×1024 feature vector through a self-attention mechanism. This yields text features.
[0024] Step 103: Based on the preprocessed growth environment parameter data, obtain the corresponding growth image information, construct an image coding network, and use the image coding network to extract features from the growth image information to obtain image features; In this embodiment, an image coding network is constructed. Growth image information obtained based on preprocessed growth environment parameter data is input into the image coding network. A first convolutional layer is performed using 32 3x3 convolutional kernels with a stride of 2 and the ReLU activation function. Subsequent layers use 64, 128, 256, and 512 convolutional kernels, each using 3x3 kernels with a stride of 2 and the ReLU activation function. With each additional layer, the feature map depth increases while its size decreases. The output feature map of the convolutional layers is then subjected to average pooling. When the feature map size is reduced to 16×16×512, a fully connected layer is connected, finally obtaining a feature vector of size 1×1024. Image features are obtained.
[0025] Step 104: Through the temperature linear fusion module, the text features and image features are fused to form a comprehensive feature representation; In this embodiment, a temperature linear fusion module is constructed, and its calculation formula is as follows:
[0026]
[0027] in, This represents the recorded daily average temperature. , , These are the daily minimum temperature, daily maximum temperature, and daily average temperature, respectively, representing the three base point temperatures at different stages of crop development.
[0028] Step 105: Use a multilayer perceptron prediction model to predict the time required for crops to reach different growth stages. Train the model by constructing a minimum mean square error loss, calculate the average of the sum of squares of the differences between the predicted value and the actual growth stage, evaluate the prediction error of the model, and adjust the model parameters using the Adam optimization algorithm.
[0029] In this embodiment, a multilayer perceptron prediction model is constructed. Each dimension of the multilayer perceptron prediction model includes an input layer, a hidden layer, and an output layer. Let the input layer be represented by a vector. If this is the case, then the output of the hidden layer is f(W). 1 +b 1 The output of the output layer is softmax(W). 2 1 +b 2 ), where W 1 b represents the weight. 1 The variable represents the bias, and the function f represents the sigmoid or tanh function. The regression from the hidden layer to the output layer is softmax regression. 1 The output f(W) of the hidden layer is represented by 1 +b 1 ); The network structure formula of a multilayer perceptron is expressed as follows:
[0030] Where G represents the softmax activation function, and h represents the number of hidden layers. and This represents the weights and biases of the i-th hidden layer. This represents the input feature vector. and represents the weights and biases of the input layer, and S represents the activation function.
[0031] In this embodiment, the loss function adopts the minimum mean square error loss, and its calculation formula is as follows:
[0032] Where m represents the number of training samples. This indicates the time from the model's predicted growth period to the corresponding growth period in the crop's textual description, where T represents the actual number of days from the crop's growth period.
[0033] Please see Figure 2 The present invention provides a schematic diagram of the structure of a crop growth period prediction system. The system includes a data annotation module, a text feature encoding module, an image feature extraction module, a feature fusion module, and a model training module. The data annotation module 201 is used to acquire growth environment parameter data of crops at different growth stages through sensors installed in the field, and to perform data annotation and preprocessing on the growth environment parameter data to obtain preprocessed growth environment parameter data. The text feature encoding module 202 is used to encode the preprocessed growth environment parameter data using a multimodal CLIP text encoder, and to convert the preprocessed growth environment parameter data into a high-dimensional vector representation to obtain text features. The image feature extraction module 203 is used to obtain the corresponding growth image information based on the preprocessed growth environment parameter data, construct an image coding network, and use the image coding network to extract features from the growth image information to obtain image features. The feature fusion module 204 is used to fuse text features and image features through the temperature linear fusion module to form a comprehensive feature representation; The model training module 205 is used to predict the time required for crops to reach different growth stages using a multilayer perceptron prediction model. The model is trained by constructing a minimum mean square error loss, calculating the average of the sum of squares of the difference between the predicted value and the actual growth stage, evaluating the prediction error of the model, and adjusting the model parameters through the Adam optimization algorithm.
[0034] In this embodiment, the preliminary processing submodule is used to obtain growth environment parameter data of crops at different growth stages, and to perform data annotation, missing value handling and standardization on the growth environment parameter data to obtain the preliminary processed growth environment parameter data. The first calculation submodule is used to set the neighborhood radius of each attribute in the pre-processed growth environment parameter data, and calculate the neighborhood relationship matrix based on the neighborhood radius; The second calculation submodule is used to calculate the neighborhood rough entropy for each attribute based on the corresponding neighborhood relationship matrix, analyze the importance of each data point in its neighborhood based on the results of the neighborhood rough entropy, and calculate the importance of each data point and its relative neighborhood. The third calculation submodule is used to calculate an outlier factor for each data point, where the outlier factor is obtained by weighted summation based on neighborhood rough entropy, importance and neighborhood relative ratio; The data point marking submodule is used to mark a data point as an outlier when its outlier factor is greater than a threshold, and add it to the outlier set. The outlier set is then removed from the growth environment parameter data to obtain the preprocessed growth environment parameter data.
[0035] In this embodiment, a construction submodule is used to construct an image coding network, which inputs growth image information obtained based on preprocessed growth environment parameter data into the image coding network. The convolutional submodule uses 32 3x3 convolutional kernels with a stride of 2 and ReLU activation function for the first layer of convolution. Then, it uses 64, 128, 256, and 512 convolutional kernels. Each layer uses 3x3 convolutional kernels with a stride of 2 and ReLU activation function. With each additional layer, the depth of the feature map increases and the size decreases. The average pooling submodule performs average pooling on the output feature maps of the convolutional layers. When the feature map size is reduced to 16×16×512, it is then connected to a fully connected layer, finally obtaining a feature vector of size 1×1024. Image features are obtained.
[0036] By implementing the above scheme, this invention can predict key time points in different growth stages of crops, prepare production materials in a timely manner, and take more scientific and standardized agronomic measures to increase crop yield.
[0037] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely preferred examples and are not intended to limit the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of the present invention is defined by the appended claims and their equivalents.
Claims
1. A method for predicting the growth period of crops, characterized in that, The method for predicting the growth period of crops includes the following steps: Data on growth environment parameters of crops at different growth stages are acquired by sensors installed in the field. The growth environment parameter data is then labeled and preprocessed to obtain preprocessed growth environment parameter data. A multimodal CLIP text encoder is used to encode the preprocessed growth environment parameter data, converting the preprocessed growth environment parameter data into a high-dimensional vector representation to obtain text features; Based on the preprocessed growth environment parameter data, the corresponding growth image information is obtained, an image coding network is constructed, and the image coding network is used to extract features from the growth image information to obtain image features. The text features and image features are fused using a temperature linear fusion module to form a comprehensive feature representation. A multilayer perceptron prediction model is used to predict the time required for crops to reach different growth stages. The model is trained by constructing a minimum mean square error loss, the average of the sum of squares of the differences between the predicted value and the actual growth stage is calculated, the prediction error of the model is evaluated, and the model parameters are adjusted by the Adam optimization algorithm.
2. The method for predicting the growth period of crops as described in claim 1, characterized in that, The process involves acquiring crop growth environment parameter data at different growth stages using sensors installed in the field, and then labeling and preprocessing this data to obtain preprocessed growth environment parameter data, including: Data on growth environment parameters of crops at different growth stages are obtained, and the growth environment parameter data is labeled, missing values are handled, and standardized to obtain pre-processed growth environment parameter data. Set the neighborhood radius for each attribute in the initially processed growth environment parameter data, and calculate the neighborhood relationship matrix based on the neighborhood radius; For each attribute, calculate the neighborhood coarse entropy based on the corresponding neighborhood relationship matrix, analyze the importance of each data point in its neighborhood based on the results of the neighborhood coarse entropy, and calculate the importance of each data point and its relative neighborhood. For each data point, an outlier factor is calculated, which is obtained by weighted summation based on neighborhood rough entropy, importance, and neighborhood relative ratio; If the outlier factor of a data point is greater than the threshold, the data point is marked as an outlier and added to the outlier set. The outlier set is then removed from the growth environment parameter data to obtain the preprocessed growth environment parameter data.
3. The method for predicting the growth period of crops as described in claim 1, characterized in that, The text encoder employing multimodal CLIP performs feature encoding on the preprocessed growth environment parameter data, converting the preprocessed growth environment parameter data into a high-dimensional vector representation to obtain text features, including: The pre-processed growth environment parameter data are arranged in order according to the time series and text labels are formed, where each set of data represents the environmental conditions at the corresponding time. The text encoder in the multimodal CLIP model is used to process the preprocessed growth environment parameter data through word embedding, position embedding and tag embedding. The word embedding, position embedding and tag embedding are added together to obtain the context representation. The context representation is transformed into a feature vector of size 1×1024 using a self-attention mechanism. This yields text features.
4. The method for predicting the growth period of crops as described in claim 1, characterized in that, The process involves obtaining corresponding growth image information based on preprocessed growth environment parameter data, constructing an image coding network, and using the image coding network to extract features from the growth image information to obtain image features, including: An image coding network is constructed, and growth image information obtained based on preprocessed growth environment parameter data is input into the image coding network. The first layer of convolution is performed using 3x3 kernels, 32 kernels, a stride of 2, and the ReLU activation function. Then, 64, 128, 256, and 512 kernels are used. Each layer uses 3x3 kernels, a stride of 2, and the ReLU activation function. With each additional layer, the depth of the feature map increases and the size decreases. The output feature map of the convolutional layer is subjected to average pooling through a pooling layer. When the feature map size is reduced to 16×16×512, a fully connected layer is then connected to obtain a feature vector of size 1×1024. Image features are obtained.
5. The method for predicting the growth period of crops as described in claim 1, characterized in that, The step involves using a temperature linear fusion module to fuse the text features and the image features to form a comprehensive feature representation, including: The temperature linear fusion module is constructed, and its calculation formula is as follows: in, This represents the recorded daily average temperature. , , These are the daily minimum temperature, daily maximum temperature, and daily average temperature, respectively, representing the three base point temperatures at different stages of crop development.
6. The method for predicting the growth period of crops as described in claim 1, characterized in that, The method of using a multilayer perceptron prediction model to predict the time required for crops to reach different growth stages includes: Construct a multilayer perceptron prediction model, wherein each dimension of the multilayer perceptron prediction model includes an input layer, a hidden layer, and an output layer. Let the input layer be represented by a vector. If this is the case, then the output of the hidden layer is f(W). 1 +b 1 The output of the output layer is softmax(W). 2 1 +b 2 ), where W 1 b represents the weight. 1 The variable represents the bias, and the function f represents the sigmoid or tanh function. The regression from the hidden layer to the output layer is softmax regression. 1 The output f(W) of the hidden layer is represented by 1 +b 1 ); The network structure formula of a multilayer perceptron is expressed as follows: Where G represents the softmax activation function, and h represents the number of hidden layers. and This represents the weights and biases of the i-th hidden layer. This represents the input feature vector. and represents the weights and biases of the input layer, and S represents the activation function.
7. The method for predicting the growth period of crops as described in claim 1, characterized in that, The process involves training the model by constructing a minimum mean squared error loss, calculating the average of the sum of squares of the differences between the predicted value and the actual reproductive period, evaluating the model's prediction error, and adjusting the model parameters using the Adam optimization algorithm, including: The loss function uses the minimum mean squared error loss, and its calculation formula is as follows: Where m represents the number of training samples. This indicates the time from the model's predicted growth period to the corresponding growth period in the crop's textual description, where T represents the actual number of days from the crop's growth period.
8. A crop growth period prediction system, characterized in that, The crop growth period prediction system includes a data annotation module, a text feature encoding module, an image feature extraction module, a feature fusion module, and a model training module. The data annotation module is used to acquire growth environment parameter data of crops at different growth stages through sensors installed in the field, and to perform data annotation and preprocessing on the growth environment parameter data to obtain preprocessed growth environment parameter data. The text feature encoding module is used to encode the preprocessed growth environment parameter data using a multimodal CLIP text encoder, converting the preprocessed growth environment parameter data into a high-dimensional vector representation to obtain text features; The image feature extraction module is used to obtain corresponding growth image information based on preprocessed growth environment parameter data, construct an image coding network, and use the image coding network to extract features from the growth image information to obtain image features. The feature fusion module is used to fuse the text features and the image features through the temperature linear fusion module to form a comprehensive feature representation; The model training module is used to predict the time required for crops to reach different growth stages using a multilayer perceptron prediction model. The model is trained by constructing a minimum mean squared error loss, calculating the average of the sum of squares of the differences between the predicted values and the actual growth stages, evaluating the prediction error of the model, and adjusting the model parameters using the Adam optimization algorithm.
9. A crop growth period prediction system as described in claim 8, characterized in that, The data annotation module includes a preliminary processing submodule, a first calculation submodule, a second calculation submodule, a third calculation submodule, and a data point marking submodule, wherein, The preliminary processing submodule is used to acquire growth environment parameter data of crops at different growth stages, and to perform data annotation, missing value handling and standardization on the growth environment parameter data to obtain the preliminary processed growth environment parameter data. The first calculation submodule is used to set the neighborhood radius of each attribute in the pre-processed growth environment parameter data, and calculate the neighborhood relationship matrix based on the neighborhood radius; The second calculation submodule is used to calculate the neighborhood rough entropy for each attribute based on the corresponding neighborhood relationship matrix, analyze the importance of each data point in its neighborhood based on the results of the neighborhood rough entropy, and calculate the importance of each data point and its relative neighborhood. The third calculation submodule is used to calculate an outlier factor for each data point, wherein the outlier factor is obtained by weighted summation of neighborhood rough entropy, importance and neighborhood relative ratio; The data point marking submodule is used to mark a data point as an outlier when its outlier factor is greater than a threshold, and add it to the outlier set. The outlier set is then removed from the growth environment parameter data to obtain the preprocessed growth environment parameter data.
10. A crop growth period prediction system as described in claim 8, characterized in that, The image feature extraction module includes a construction submodule, a convolution submodule, and an average pooling submodule, wherein... A submodule is constructed to build an image coding network, which inputs growth image information obtained based on preprocessed growth environment parameter data into the image coding network. The convolutional submodule uses 32 3x3 convolutional kernels with a stride of 2 and ReLU activation function for the first layer of convolution. Then, it uses 64, 128, 256, and 512 convolutional kernels. Each layer uses 3x3 convolutional kernels with a stride of 2 and ReLU activation function. With each additional layer, the depth of the feature map increases and the size decreases. The average pooling submodule performs average pooling on the output feature maps of the convolutional layers. When the feature map size is reduced to 16×16×512, it is then connected to a fully connected layer, finally obtaining a feature vector of size 1×1024. Image features are obtained.