Crop monitoring method based on remote sensing land utilization and land change data

Abstract

The invention discloses a crop monitoring method based on remote sensing land utilization and land change data, which realizes macroscopic crop information capture and data preprocessing of a large-range farmland area through remote sensing satellite image data acquisition, effectively removes noise, corrects distortion and enhances features, provides a high-quality basis for classification and identification, and improves the quality of crops. The customized model keeps the feature extraction capability and adapts to the crop classification task by freezing the convolutional layer of the pre-training model and adding a new layer, and uses an optimizer and a loss function compilation model suitable for multi-class classification to ensure the training stability and convergence, so as to improve the crop classification efficiency. A large amount of preprocessed image data is trained to improve the generalization ability and classification precision of the model, the crop classification model can process remote sensing images in real time or quasi-real time, crop type information is output, intelligent and automatic monitoring, result analysis and output are achieved, and a visual distribution diagram and a detailed monitoring report are generated. And powerful data support is provided for agricultural production decision making.

Description

Technical Field

[0001] The present invention relates to the technical field of crop monitoring, and in particular, to a crop monitoring method based on remote sensing land use and land change data. Background Art

[0002] Crop planting is an important link in agricultural production. Scientific and efficient monitoring of crop planting is crucial for improving agricultural production efficiency and ensuring food security. However, traditional agricultural survey methods mainly rely on manual on-site surveys, which are time-consuming and laborious, and there are uncertainties. In addition, with the transformation of agriculture from traditional decentralized household management to large-scale, intensive and informatized management, the demand for large-scale, rapid, timely and dynamic spatial information is becoming increasingly urgent. Therefore, it is particularly important to find a method that can efficiently and accurately obtain crop planting information. Summary of the Invention

[0003] In view of this, the present invention proposes a crop monitoring method based on remote sensing land use and land change data, which can effectively solve the defects of the prior art that rely on manual on-site surveys, are time-consuming and laborious, and have uncertainties.

[0004] The technical solution of the present invention is realized as follows:

[0005] A crop monitoring method based on remote sensing land use and land change data specifically includes:

[0006] Data acquisition: Obtain remote sensing satellite image data containing crop information;

[0007] Data preprocessing: Perform preprocessing of denoising, calibration and enhancement on the remote sensing satellite image data;

[0008] Model selection: Select at least one pre-trained convolutional neural network model as the basic model for crop classification;

[0009] Model customization: Freeze the convolutional base layer of the pre-trained model, and add a new fully connected layer and a classification layer thereon to obtain a customized model to adapt to the crop classification task;

[0010] Model compilation: Compile the customized model using an optimizer and a loss function suitable for multi-class classification;

[0011] Model training: Use the preprocessed remote sensing satellite image data to train the customized model to obtain a crop classification model;

[0012] Crop recognition: Input the remote sensing satellite image data to be processed into the crop classification model, and the crop classification model outputs the crop type;

[0013] Result analysis and output: Analyze the classification results, generate a crop distribution map and a monitoring report, and output the results.

[0014] As a further optional solution of the crop monitoring method based on remote sensing land use and land change data, freeze the convolutional base layer of the pre-trained model and add a new fully connected layer and a classification layer on it to adapt to the crop classification task, specifically including:

[0015] Load the pre-trained model and keep the weights of its convolutional base layer unchanged;

[0016] Set the parameters of the convolutional base layer to be untrainable, and remove the last average pooling layer and the original fully connected layer from the pre-trained model to obtain the frozen convolutional base layer;

[0017] Add a new fully connected layer on top of the frozen convolutional base layer to map the features extracted by the convolutional layer to the classification space;

[0018] According to the specific number of categories of the crop classification task, set the number of output neurons of the classification layer, and use the softmax activation function to output the prediction probability of each category to obtain a new classification layer;

[0019] Connect the frozen convolutional base layer, the new fully connected layer and the classification layer to build a customized model.

[0020] As a further optional solution of the crop monitoring method based on remote sensing land use and land change data, compile the customized model using an optimizer and a loss function suitable for multi-class classification, specifically including:

[0021] Before starting each training epoch, set the model to training mode;

[0022] Before the training loop, initialize a variable to track the defined loss function;

[0023] Use the data loader to iterate over the training data;

[0024] Before each batch starts, zero the gradients of all variables to be optimized;

[0025] Pass the input data to the model and obtain the output of the model;

[0026] Use the defined loss function to calculate the loss between the model output and the true labels;

[0027] Calculate the gradients of the loss with respect to all trainable parameters;

[0028] Use the optimizer to adjust the values of the model parameters according to the calculated gradients.

[0029] As a further alternative of the crop monitoring method based on remote sensing land use and land change data, the defined loss function is the focal cross-entropy loss function, specifically:

[0030]

[0031] where y is the one-hot encoding of the true label, and yi represents the true label of the i-th category;

[0032] is the probability distribution predicted by the model, indicating the predicted probability of the i-th category;

[0033] C is the total number of categories;

[0034] α is the weight balancing factor, used to adjust the weights between different categories;

[0035] γ is the focusing adjustment factor, used to adjust the degree of attention to difficult-to-classify samples.

[0036] As a further alternative of the crop monitoring method based on remote sensing land use and land change data, the analysis of the classification results to generate a crop distribution map and a monitoring report specifically includes:

[0037] Collect the geographical location data of the crops according to the classification results;

[0038] Use GIS software to import the geographical location data and the classification results;

[0039] According to the classification results, set different colors or icons for different growth conditions to generate a crop distribution map;

[0040] Combined with the crop distribution map, mark the growth conditions and their reasons in each area to generate a monitoring report.

[0041] A crop monitoring system based on remote sensing land use and land change data includes:

[0042] A data acquisition module for acquiring remote sensing satellite image data containing crop information;

[0043] A data preprocessing module for preprocessing the remote sensing satellite image data by denoising, calibration, and enhancement;

[0044] A model selection module for selecting at least one pre-trained convolutional neural network model as the basic model for crop classification;

[0045] A model customization module for freezing the convolutional base layer of the pre-trained model and adding new fully connected layers and classification layers thereon to obtain a customized model adapted to the crop classification task;

[0046] A model compilation module that compiles the customized model using an optimizer and a loss function suitable for multi-class classification;

[0047] A model training module that trains the customized model using the preprocessed remote sensing satellite image data to obtain a crop classification model;

[0048] A crop recognition module that inputs the remote sensing satellite image data to be processed into the crop classification model, and the crop classification model outputs the crop type;

[0049] A result analysis and output module that analyzes the classification results, generates a crop distribution map and a monitoring report, and outputs the results.

[0050] A computing device includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the steps of any one of the above-mentioned crop monitoring methods based on remote sensing land use and land change data are implemented.

[0051] A computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps of any one of the above-mentioned crop monitoring methods based on remote sensing land use and land change data are implemented.

[0052] The beneficial effects of the present invention are as follows: Through the acquisition of remote sensing satellite image data, it is possible to quickly cover a large range of farmland areas and capture crop information at the macroscopic level. The data preprocessing steps effectively remove image noise, correct image distortion, and enhance image features, providing a high-quality data basis for subsequent classification and recognition. By freezing the convolutional base layer of the pre-trained model and adding new fully connected layers and classification layers on this basis, the customization of the model is realized. This method not only retains the feature extraction ability of the pre-trained model but also can be adaptively adjusted for the crop classification task. The customized model can better learn the unique image features of crops, thereby improving the accuracy and robustness of classification. Compiling the customized model using an optimizer and a loss function suitable for multi-class classification ensures the stability and convergence of the model during training. By training with a large amount of preprocessed remote sensing satellite image data, the model can learn the image features of crops at different growth stages and different environmental conditions, further improving the generalization ability and classification accuracy of the model. The crop classification model can process the remote sensing satellite image data to be processed in real-time or near real-time and output crop type information, realizing the intelligence and automation of crop monitoring. The result analysis and output steps can generate an intuitive crop distribution map and a detailed monitoring report, providing strong data support for agricultural production decision-making. Description of the Drawings

[0053] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the accompanying drawings required for the description of the embodiments or the prior art. Obviously, the accompanying drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other accompanying drawings can be obtained based on these drawings.

[0054] Figure 1 It is a schematic flowchart of a crop monitoring method based on remote sensing land use and land change data of the present invention;

[0055] Figure 2 It is a schematic diagram of the composition of a crop monitoring system based on remote sensing land use and land change data of the present invention;

[0056] Figure 3 It is a schematic diagram of the composition of a computing device of the present invention. Specific Embodiments

[0057] The following will clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only some of the embodiments of the present invention, rather than all of them. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts fall within the scope of protection of the present invention.

[0058] Refer to Figures 1 to 3 , a crop monitoring method based on remote sensing land use and land change data specifically includes:

[0059] Data Acquisition: Obtain remote sensing satellite image data containing crop information;

[0060] Data Preprocessing: Perform preprocessing of denoising, calibration, and enhancement on the remote sensing satellite image data;

[0061] Model Selection: Select at least one pre-trained convolutional neural network model as the basic model for crop classification;

[0062] Model Customization: Freeze the convolutional base layer of the pre-trained model and add new fully connected layers and classification layers thereon to obtain a customized model to adapt to the crop classification task;

[0063] Model Compilation: Compile the customized model using an optimizer and loss function suitable for multi-class classification;

[0064] Model Training: Use the preprocessed remote sensing satellite image data to train the customized model to obtain a crop classification model;

[0065] Crop Recognition: Input the remote sensing satellite image data to be processed into the crop classification model, and the crop classification model outputs the crop type;

[0066] Result Analysis and Output: Analyze the classification results, generate a crop distribution map and a monitoring report, and output the results.

[0067] In this embodiment, through the acquisition of remote sensing satellite image data, it is possible to quickly cover a large range of farmland areas, realize the capture of crop information at the macroscopic level. The data preprocessing steps effectively remove image noise, correct image distortion, and enhance image features, providing a high-quality data basis for subsequent classification and recognition. By freezing the convolutional base layer of the pre-trained model and adding new fully connected layers and classification layers on this basis, the customization of the model is realized. This method not only retains the feature extraction ability of the pre-trained model but also can be adaptively adjusted for the crop classification task. The customized model can better learn the unique image features of crops, thereby improving the accuracy and robustness of classification. Compiling the customized model using an optimizer and loss function suitable for multi-class classification ensures the stability and convergence of the model during training. By training with a large amount of preprocessed remote sensing satellite image data, the model can learn the image features of crops at different growth stages and different environmental conditions, further enhancing the generalization ability and classification accuracy of the model. The crop classification model can process the remote sensing satellite image data to be processed in real-time or near real-time and output crop type information, realizing the intelligence and automation of crop monitoring. The result analysis and output steps can generate an intuitive crop distribution map and a detailed monitoring report, providing strong data support for agricultural production decision-making.

[0068] Preferably, freezing the convolutional base layer of the pre-trained model and adding new fully connected layers and classification layers thereon to adapt to the crop classification task specifically includes:

[0069] Load the pre-trained model and keep the weights of its convolutional base layer unchanged;

[0070] Set the parameters of the convolutional base layer to be non-trainable, and remove the last average pooling layer and the original fully connected layer from the pre-trained model to obtain the frozen convolutional base layer;

[0071] Add a new fully connected layer on top of the frozen convolutional base layer to map the features extracted by the convolutional layer to the classification space;

[0072] According to the specific number of categories of the crop classification task, set the number of output neurons of the classification layer, and use the softmax activation function to output the prediction probability of each category to obtain a new classification layer;

[0073] Connect the frozen convolutional base layer, the new fully connected layer, and the classification layer to build a customized model.

[0074] In this embodiment, knowledge transfer in deep learning is achieved by loading a pre-trained model and keeping the weights of its convolutional base layer unchanged. The pre-trained model has learned rich feature representations on a large amount of image data, and these features are also valuable for the crop classification task. Freezing the convolutional base layer means that these already learned feature representations will not be modified during subsequent training, thus ensuring the effective reuse of features and avoiding the huge computational resources and time costs required for training the model from scratch; removing the last average pooling layer and the original fully connected layer of the pre-trained model, and adding a new fully connected layer and a classification layer on top of the frozen convolutional base layer. This step enables the model to be adaptively adjusted according to the specific requirements of the crop classification task. The new fully connected layer can learn the classification features unique to crops based on the features extracted by the convolutional base layer, thereby improving the adaptability of the model to the crop classification task; setting the number of output neurons of the classification layer according to the specific number of categories in the crop classification task, and using the softmax activation function to output the prediction probability for each category. This step ensures that the model can accurately classify crops. The Softmax activation function can convert the output of the fully connected layer into a probability distribution, enabling the model to output the prediction probability for each category, which is convenient for subsequent result analysis and output; since the weights of the convolutional base layer are frozen, only the weights of the newly added fully connected layer and classification layer need to be updated during training, which greatly reduces the number of parameters to be trained, thereby improving the training efficiency of the model. In addition, since the pre-trained model has learned rich feature representations, the newly added layers can converge to the optimal solution faster, further shortening the training time of the model; by freezing the convolutional base layer of the pre-trained model and adding new layers on it to build a customized model, this method combines the generalization ability of the pre-trained model and the adaptability of the newly added layers, thereby enhancing the generalization ability of the entire model. This means that the customized model can not only perform well on the training data but also accurately classify unseen crop images.

[0075] Preferably, compiling the customized model using an optimizer and a loss function suitable for multi-class classification specifically includes:

[0076] Before starting each training epoch, set the model to training mode;

[0077] Before the training loop, initialize a variable to track the defined loss function;

[0078] Use a data loader to iterate over the training data;

[0079] Before the start of each batch, zero out the gradients of all optimized variables;

[0080] Pass the input data to the model and obtain the model's output;

[0081] Use the defined loss function to calculate the loss between the model output and the true labels;

[0082] Calculate the gradients of the loss with respect to all trainable parameters;

[0083] Use the optimizer to adjust the values of the model parameters according to the calculated gradients.

[0084] In this embodiment, before starting each training cycle, the model is set to the training mode. This step ensures that all trainable parameters in the model can be updated properly during training. By correctly setting the training mode, errors or performance degradation caused by improper model configuration during training can be avoided. Initialize a variable to track the defined loss function, so that at the end of each training cycle or batch, the average loss of the model can be conveniently calculated and output. This helps monitor the training progress and performance changes of the model. By continuously tracking the loss value, problems (such as overfitting, underfitting, etc.) during the model training process can be detected in a timely manner, and corresponding measures can be taken for adjustment. Use the data loader to iterate through the training data, which can efficiently load and preprocess the data, and at the same time support batch operations. This helps reduce the I / O overhead during model training and improve the training speed. The data loader can also provide functions such as random shuffling of data and adjustment of batch size, which helps enhance the generalization ability of the model. Before the start of each batch, clearing the gradients of all variables to be optimized is a key step to ensure that the optimizer can correctly calculate and update the model parameters. If the gradients are not cleared, the gradients of the previous batch will accumulate to the current batch, resulting in inaccurate update of the model parameters. By clearing the gradients, it can be ensured that each batch is independently calculated for gradients and updated for parameters, thus avoiding errors caused by gradient accumulation. After passing the input data to the model and obtaining the output of the model, use the defined loss function to calculate the loss between the model output and the true label. This step is the core of model training, which measures the gap between the current performance of the model and the target performance. By calculating the gradients of the loss with respect to all trainable parameters, it can guide the optimizer on how to adjust the model parameters to reduce the loss value. Gradient backpropagation is one of the key steps in the deep learning training process, which realizes the parameter update from the output layer to the input layer. Use the optimizer to adjust the values of the model parameters according to the calculated gradients, which is the last step in the model training process. The optimizer (such as SGD, Adam, etc.) is responsible for updating the model parameters according to the gradient information, thereby gradually optimizing the model performance. By continuously iterating through the training data and updating the model parameters, the model can gradually learn the distribution law and feature representation of the data, thereby improving the accuracy and generalization ability for the crop classification task.

[0085] Preferably, the defined loss function is the focal cross-entropy loss function, specifically:

[0086]

[0087] where y is the one-hot encoding of the true label, and yi represents the true label of the i-th class;

[0088] is the probability distribution predicted by the model, represents the predicted probability of the i-th class;

[0089] C is the total number of categories;

[0090] α is the weight balance factor, which is used to adjust the weights between different categories. For the case of class imbalance, the minority classes can be given greater weights by setting α;

[0091] γ is the focus adjustment factor, which is used to adjust the degree of attention to difficult-to-classify samples. When γ increases, the loss contribution of the model to easy-to-classify samples will decrease, while the loss contribution to difficult-to-classify samples will increase.

[0092] In this embodiment, by introducing the weight balance factor α, the focal cross-entropy loss function can assign different weights to samples of different categories, which is particularly important when dealing with class-imbalanced datasets. It can make the model pay more attention to those minority classes or difficult-to-learn classes during the training process, thereby improving the overall classification performance. The introduction of the focus adjustment factor γ enables the loss function to adjust the degree of attention to difficult-to-classify samples. Specifically, when the value of γ is large, the model will pay more attention to those samples with lower prediction probabilities (i.e., difficult to classify). By increasing the contribution of these samples to the loss, it prompts the model to continuously optimize its classification ability for these samples during the training process. Since the focal cross-entropy loss function can balance the weights of different categories and pay attention to difficult-to-classify samples, this helps to improve the robustness of the model on complex and diverse datasets. The model can better learn the internal laws and features of the data, and thus maintain good classification performance in different scenarios. Using the focal cross-entropy loss function as the loss function can optimize the training process of the model. By reducing the contribution of easy-to-classify samples to the loss, it makes the model more focused on the learning of difficult-to-classify samples during the training process, thereby accelerating the training process and improving the training efficiency. By paying special attention to difficult-to-classify samples, the focal cross-entropy loss function helps to improve the accuracy of the model in classification tasks. The model can more accurately identify samples of different categories, reduce the situation of misclassification, and thus improve the overall classification effect.

[0093] Preferably, the analysis of the classification results to generate a crop distribution map and a monitoring report specifically includes:

[0094] Collect the geographical location data of the crops according to the classification results;

[0095] Import the geographical location data and the classification results using GIS software;

[0096] According to the classification results, set different colors or icons for different growth conditions to generate a crop distribution map;

[0097] Combined with the crop distribution map, mark the growth conditions and their reasons in each area to generate a monitoring report.

[0098] In this embodiment, collecting the geographical location data of crops according to the classification results ensures the accuracy and pertinence of the data. By combining the classification results and geographical location information, the growth conditions of different crops at different geographical locations can be accurately tracked. Using GIS (Geographic Information System) software to import the geographical location data and classification results can visually display the distribution of crops. The powerful spatial analysis ability and visualization function of GIS software enable users to clearly see the distribution of crops in different regions and their spatial relationships. According to the classification results, different colors or icons are set for different growth conditions to generate a crop distribution map. This intuitive visualization presentation method enables users to see at a glance the growth conditions of crops, including health, pests and diseases, malnutrition, etc. This helps users quickly identify problem areas and take corresponding management measures. Combining the crop distribution map, marking the growth conditions and their reasons in each region to generate a monitoring report. This step not only provides intuitive image information but also combines in-depth analysis and interpretation, enabling users to more comprehensively understand the growth conditions of crops and the reasons behind them. The monitoring report can also provide targeted suggestions and measures to help users optimize crop management strategies. Users can quickly obtain information on the distribution and growth conditions of crops, providing strong support for decision-making. At the same time, presenting the data in a visual way reduces the threshold of data interpretation and improves the efficiency of decision-making. In addition, the generation of the monitoring report also simplifies the process of data analysis and report writing, further improving work efficiency.

[0099] A crop monitoring system based on remote sensing land use and land change data, comprising:

[0100] A data acquisition module for acquiring remote sensing satellite image data containing crop information;

[0101] A data preprocessing module for preprocessing the remote sensing satellite image data by denoising, correcting and enhancing;

[0102] A model selection module for selecting at least one pre-trained convolutional neural network model as the basic model for crop classification;

[0103] A model customization module for freezing the convolutional base layer of the pre-trained model and adding new fully connected layers and classification layers thereon to obtain a customized model adapted to the crop classification task;

[0104] A model compilation module for compiling the customized model using an optimizer and loss function suitable for multi-class classification;

[0105] A model training module for training the customized model using the preprocessed remote sensing satellite image data to obtain a crop classification model;

[0106] The crop recognition module inputs the remote sensing satellite image data to be processed into the crop classification model, and the crop classification model outputs the crop types;

[0107] The result analysis and output module analyzes the classification results, generates a crop distribution map and a monitoring report, and outputs the results.

[0108] A computing device includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the steps of any one of the above-mentioned crop monitoring methods based on remote sensing land use and land change data are implemented.

[0109] A computer-readable storage medium stores a computer program. When the computer program is executed by a processor, the steps of any one of the above-mentioned crop monitoring methods based on remote sensing land use and land change data are implemented.

[0110] The above are only the preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A crop monitoring method based on remote sensing land use and land change data, characterized in that: Specifically include: Data acquisition: Acquire remote sensing satellite image data containing crop information; Data preprocessing: performing denoising, correction and enhancement preprocessing on the remote sensing satellite image data; Model selection: Select at least one pre-trained convolutional neural network model as the base model for crop classification; Model customization: Freeze the convolutional base layer of the pre-trained model and add new fully connected layers and classification layers on it to obtain a customized model to adapt to the crop classification task; Model compilation: compile the custom model using an optimizer and loss function suitable for multi-class classification; Model training: using the preprocessed remote sensing satellite image data to train the customized model to obtain a crop classification model; Crop identification: The remote sensing satellite image data to be processed is input into the crop classification model, and the crop classification model outputs the crop type; Result analysis and output: Analyze the classification results, generate crop distribution maps and monitoring reports, and output the results.

2. A crop monitoring method based on remote sensing land use and land change data according to claim 1, characterized in that: The freezing of the convolutional base layer of the pre-trained model and adding a new fully connected layer and a classification layer thereon to adapt to the crop classification task specifically includes: Load the pre-trained model and keep the weights of its convolutional base layer unchanged; Set the parameters of the convolutional base layer to non-trainable, and remove the last average pooling layer and the original fully connected layer from the pre-trained model to obtain a frozen convolutional base layer; On top of the frozen convolutional base layer, a new fully connected layer is added to map the features extracted by the convolutional layer to the classification space; According to the specific number of categories of the crop classification task, the number of output neurons in the classification layer is set, and the softmax activation function is used to output the predicted probability of each category to obtain a new classification layer; Connect the frozen convolutional base layers, new fully connected layers, and classification layers to build a customized model.

3. A crop monitoring method based on remote sensing land use and land change data according to claim 2, characterized in that: Compiling the custom model using an optimizer and loss function suitable for multi-class classification specifically includes: Before starting each training cycle, set the model to training mode; Before the training loop, initialize a variable to keep track of the defined loss function; Use the data loader to iterate over the training data; Before each batch starts, clear the gradients of all optimized variables; Pass input data to the model and get the model's output; Use the defined loss function to calculate the loss between the model output and the true label; Compute the gradient of loss with respect to all trainable parameters; Use an optimizer to adjust the values ​​of the model parameters based on the calculated gradients.

4. The crop monitoring method based on remote sensing land use and land change data according to claim 3, characterized in that: The loss function defined is the focal cross entropy loss function, specifically: Among them, y is the one-hot encoding of the true label, y i Represents the true label of the i-th category; is the probability distribution predicted by the model, represents the predicted probability of the i-th category; C is the total number of categories; α is the weight balancing factor, which is used to adjust the weights between different categories; γ is a focus adjustment factor, which is used to adjust the degree of attention paid to difficult-to-classify samples.

5. The crop monitoring method based on remote sensing land use and land change data according to claim 4, characterized in that: The classification results are analyzed to generate a crop distribution map and a monitoring report, specifically including: Collect geographical location data of crops based on the classification results; Use GIS software to import geographic location data and classification results; According to the classification results, different colors or icons are set for different growth conditions to generate crop distribution maps; Combined with the crop distribution map, the growth conditions and reasons of each area are marked to generate a monitoring report.

6. A crop monitoring system based on remote sensing land use and land change data, characterized in that: include: A data acquisition module, used for acquiring remote sensing satellite image data containing crop information; A data preprocessing module, for preprocessing the remote sensing satellite image data by denoising, correcting and enhancing; A model selection module, used for selecting at least one pre-trained convolutional neural network model as a basic model for crop classification; A model customization module, used for freezing the convolution base layer of the pre-trained model and adding a new fully connected layer and a classification layer thereto to obtain a customized model, wherein the customized model is suitable for a crop classification task; A model compilation module, compiling the custom model using an optimizer and loss function suitable for multi-class classification; A model training module, using the pre-processed remote sensing satellite image data to train the customized model to obtain a crop classification model; The crop identification module inputs the remote sensing satellite image data to be processed into the crop classification model, and the crop classification model outputs the crop type; The result analysis and output module analyzes the classification results, generates crop distribution maps and monitoring reports, and outputs the results.

7. A computing device, characterized in that The method comprises a memory, a processor and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the steps of the crop monitoring method based on remote sensing land use and land change data described in any one of claims 1 to 5 are implemented.

8. A computer-readable storage medium, characterized in that: The storage medium stores a computer program, which, when executed by a processor, implements the steps of the crop monitoring method based on remote sensing land use and land change data as recited in any one of claims 1 to 5.

Images