Deep learning-based rice planting structure remote sensing extraction method
Through the remote sensing extraction method of rice planting structure based on deep learning, the problem of difficulty in accurately extracting rice planting structure and pest information in the existing technology is solved, and higher extraction accuracy and efficiency are achieved, providing a more accurate basis for agricultural management.
Patent Information
- Application Number
- CN202411899867.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-12-23
- Publication Date
- 2025-04-15
- Estimated Expiration
- Not applicable · inactive patent
AI Technical Summary
It is difficult for the prior art to accurately extract rice planting structure and pest information, affecting yield estimates and pest evaluations.
The remote sensing extraction method of rice planting structure based on deep learning is adopted. By obtaining multi-time phase satellite remote sensing image data, a multi-scale image pyramid and deep learning classification model are constructed, and the standard rice planting structure characteristics are combined for training and classification, and rice planting structure and pest information are extracted.
It improves the accuracy and efficiency of rice planting area extraction, provides a more accurate scientific basis for agricultural management and decision-making, and enhances the model's adaptability to complex terrain and diversified farmland environments.
Smart Images
Figure CN119832425A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of rice planting, and in particular to a remote sensing extraction method of rice planting structure based on deep learning. Background Art
[0002] Rice is an herbaceous plant grown in paddy fields. It belongs to the annual grass family and is mainly divided into two categories: japonica rice and indica rice. It has the growth characteristics of short-day, thermophilic and semi-aquatic. In the growing season, the shorter the daylight hours, the earlier the flowering and fruiting. The minimum temperature for seed germination is 12-14℃. Although it needs more water, it will grow poorly and have low yields if it is flooded or lacks water all the time. As one of the most important food crops in the world, rice is an important grain in the human diet, providing the main food source for billions of people. It can be processed into rice, cooked rice, vermicelli, cake, flour and other foods. In addition, rice also has a wide range of application value in medicine, feed, industry, energy and ecological services.
[0003] Accurate extraction of rice planting structure is of great significance to rice yield estimation and evaluation of pest and disease qualified rate. Under the joint action of multiple factors such as climate change, regional rice planting structure is constantly changing. Rice can be subdivided into double-season rice, single-season late rice, medium-season rice, regenerated rice and other subdivided types according to different rice-growing systems. How to distinguish them is a problem to be solved. Remote sensing technology has the advantages of wide coverage, short detection cycle, strong timeliness and low cost. It is an important means to extract rice planting structure. How to use remote sensing technology, combined with deep learning technology, to obtain the distribution of different rice structures and the judgment of pests and diseases is a problem to be solved. Therefore, the present invention proposes a rice planting structure remote sensing extraction method based on deep learning to solve the problems existing in the prior art. Summary of the invention
[0004] In response to the above problems, the present invention proposes a remote sensing extraction method for rice planting structure based on deep learning. The remote sensing extraction method for rice planting structure based on deep learning improves the accuracy and efficiency of rice planting area extraction, provides a more accurate scientific basis for agricultural management and decision-making, and combines the powerful feature extraction capability of deep learning with the advantages of multi-scale fusion, thereby enhancing the model's adaptability to complex terrain and diversified farmland environment.
[0005] To achieve the purpose of the present invention, the present invention is implemented by the following technical scheme: a rice planting structure remote sensing extraction method based on deep learning, comprising the following steps:
[0006] S1: Obtain multi-temporal satellite remote sensing image data of rice areas, pre-process the data, synthesize bands, and calculate spectral indices;
[0007] S2: Construct a multi-scale image pyramid, obtain remote sensing images at different scales, extract features at each scale, and fuse them to obtain a multi-scale feature fusion map;
[0008] S3: Build a deep learning classification model, take the characteristics of the standard rice planting structure as input, and train the deep learning model once;
[0009] S4: Use the trained deep learning model to classify the multi-scale feature fusion map, obtain the classification results of rice planting structure, and perform area measurement;
[0010] S5: In the deep learning classification model, standard feature parameters of color, texture and shape frameworks for different classifications of rice are input for secondary training;
[0011] S6: Use the trained deep learning model to compare the classified rice planting structures, obtain the qualified rate of each type of rice, and determine the abnormal structure area;
[0012] S7: Construct a rice pest and disease ontology knowledge base, convert it into standard structure parameters, and input it into a deep learning classification model for three trainings;
[0013] S8: Use the trained deep learning model to classify the abnormal structure area and obtain the abnormal type classification results of different abnormal structure rice in the abnormal structure area.
[0014] A further improvement is that: S1 comprises the following steps:
[0015] Obtain multi-temporal and multi-spectral satellite remote sensing image data of rice areas, including visible light, near infrared and short-wave infrared bands;
[0016] Preprocess the remote sensing image data, perform radiation calibration, atmospheric correction, geometric correction and noise removal in sequence;
[0017] The preprocessed remote sensing image data was synthesized, the NDVI normalized difference vegetation index was calculated, and the characteristic information related to rice growth was extracted.
[0018] A further improvement is that: S2 comprises the following steps:
[0019] Construct a multi-scale image pyramid to divide remote sensing image data into images of different scales, including 7px×7px, 13px×13px, and 19px×19px;
[0020] At each scale, convolutional neural network (CNN) is used to extract features and obtain feature maps of different scales.
[0021] Feature maps of different scales are fused by weighted average or maximum fusion to form a multi-scale feature fusion map.
[0022] A further improvement is that: S3 comprises the following steps:
[0023] Build a deep learning classification model, using the ResNet model or the DenseNet model;
[0024] The deep learning model is trained once using the features of the standard rice planting structure as input;
[0025] The learning rate is set to 0.001, the batch size is set to 32, the number of iterations is set to 100, and the cross-validation method is used to optimize the model.
[0026] A further improvement is that: said S4 comprises the following steps:
[0027] The trained deep learning model is used to classify the multi-scale feature fusion map to obtain the classification layout of rice planting structure;
[0028] Morphological filtering and connected domain analysis are used to remove noise, fill holes, and optimize edge processing of the classification layout;
[0029] The number of pixels of the rice planting structure in each classification layout is counted, and the rice planting area of each classification layout is calculated according to the preset ratio classification.
[0030] A further improvement is that: S5 comprises the following steps:
[0031] Collect standard feature parameters of color, texture and shape frameworks for different classifications of rice from relevant knowledge bases to form a data set;
[0032] The standard characteristic parameters of each category cover the rice structural variables of this category of rice at different growth stages and under different environmental conditions;
[0033] In the deep learning classification model, ResNet was selected as the model architecture for rice feature extraction and classification, and the model was adjusted to add convolutional layers;
[0034] The data set is input into the deep learning model for secondary training, using cross-validation and early stopping strategies to avoid model overfitting.
[0035] A further improvement is that: S6 comprises the following steps:
[0036] The trained deep learning model is used to compare the classified rice planting structures separately, and the rice planting structures of each category are further refined based on the color, texture and shape framework meeting the standard feature parameters;
[0037] Determine the ratio of qualified rice structures to the rice area of each type of rice planting structure layout, and determine its qualified rate;
[0038] By using the spatial information of remote sensing images, abnormal areas are marked on the images to form abnormal structural areas.
[0039] A further improvement is that: said S7 comprises the following steps:
[0040] Using external sources, expert databases, text databases and corresponding traditional knowledge bases as data sources, we build a rice pest and disease ontology knowledge base and convert it into specific color, texture and shape framework structure parameters for different classifications of rice.
[0041] The parameters of the above-mentioned different classifications of rice were input into the deep learning classification model, and cross-validation was used to perform three trainings corresponding to the abnormal structural regions of the different classifications of rice.
[0042] A further improvement is that: said S8 comprises the following steps:
[0043] Using the forward propagation of the trained deep learning model, the abnormal structural area of each classified rice is reclassified;
[0044] Determine the types and distribution of different pests and diseases in each classified rice abnormal structure zone.
[0045] A further improvement is that in S8, the final extraction result is output to the relevant expert website to obtain solutions for the abnormal area.
[0046] The beneficial effects of the present invention are:
[0047] 1. The present invention obtains multi-temporal satellite remote sensing image data of rice areas, obtains remote sensing images at different scales based on a multi-scale image pyramid, fuses them to obtain a multi-scale feature fusion map, constructs a deep learning classification model, and performs a training based on the characteristics of the standard rice planting structure. The multi-scale feature fusion map is classified to obtain the classification result of the rice planting structure, and the area is calculated, which improves the accuracy and efficiency of rice planting area extraction, provides a more accurate scientific basis for agricultural management and decision-making, combines the powerful feature extraction capability of deep learning with the advantages of multi-scale fusion, and enhances the model's adaptability to complex terrain and diversified farmland environments.
[0048] 2. The present invention performs secondary training based on the classification results of rice planting structure obtained from the first training with standard feature parameters of color, texture and shape framework of rice of different classifications. Thus, rice of different classification structures are compared, and the abnormal structure area of each type of rice structure is obtained based on the standard feature parameters as the threshold, so as to facilitate the determination of the specific distribution area of planting problems and provide a basis for the overall rice quality evaluation.
[0049] 3. The present invention uses the rice pest and disease ontology knowledge base as the standard structure parameter based on the abnormal structure area obtained by the secondary training, and inputs the model for three trainings, thereby classifying the abnormal structure area, and obtaining the specific categories and distribution of pests and diseases of rice with different abnormal structures in the abnormal structure area, which is convenient for targeted remediation and solving the problem. BRIEF DESCRIPTION OF THE DRAWINGS
[0050] Figure 1 It is the front view of the present invention. DETAILED DESCRIPTION
[0051] In order to deepen the understanding of the present invention, the present invention will be further described in detail below in conjunction with examples. The examples are only used to explain the present invention and do not constitute a limitation on the protection scope of the present invention.
[0052] Embodiment 1
[0053] according to Figure 1 As shown, this embodiment proposes a rice planting structure remote sensing extraction method based on deep learning, comprising the following steps:
[0054] Obtain multi-temporal satellite remote sensing image data of rice areas, pre-process the data, synthesize bands, and calculate spectral indices;
[0055] Construct a multi-scale image pyramid, obtain remote sensing images at different scales, extract features of each scale, and fuse them to obtain a multi-scale feature fusion map;
[0056] Build a deep learning classification model, take the characteristics of the standard rice planting structure as input, and train the deep learning model once;
[0057] The trained deep learning model is used to classify the multi-scale feature fusion map, obtain the classification results of rice planting structure, and perform area measurement;
[0058] In the deep learning classification model, standard feature parameters for the color, texture, and shape frameworks of different rice classifications were input for secondary training;
[0059] The trained deep learning model is used to compare the classified rice planting structures, obtain the qualified rate of each type of rice, and determine the abnormal structure area;
[0060] Construct a rice pest and disease ontology knowledge base, convert it into standard structure parameters, and input it into a deep learning classification model for three trainings;
[0061] The trained deep learning model is used to classify the abnormal structure areas, and the classification results of abnormal types of rice with different abnormal structures in the abnormal structure areas are obtained.
[0062] The present invention combines the powerful feature extraction capability of deep learning with the advantages of multi-scale fusion, improves the accuracy and efficiency of rice planting area extraction, helps to improve the monitoring and management level of rice production, and contributes to the sustainable development of agriculture.
[0063] Embodiment 2
[0064] according to Figure 1 As shown, this embodiment proposes a rice planting structure remote sensing extraction method based on deep learning, comprising the following steps:
[0065] S1: Acquire multi-temporal satellite remote sensing image data of the rice area, perform preprocessing and band synthesis, and calculate spectral index on the data; specifically, the following steps are included: acquire multi-temporal and multi-spectral satellite remote sensing image data of the rice area, including visible light, near infrared and short-wave infrared bands; preprocess the remote sensing image data, perform radiation calibration, atmospheric correction, geometric correction and noise removal in sequence; perform band synthesis on the preprocessed remote sensing image data, calculate the NDVI normalized difference vegetation index, and extract characteristic information related to rice growth.
[0066] S2: Construct a multi-scale image pyramid, obtain remote sensing images at different scales, extract features at each scale, and fuse them to obtain a multi-scale feature fusion map; specifically, the following steps are included: construct a multi-scale image pyramid, divide the remote sensing image data into images of different scales, including 7px×7px, 13px×13px, and 19px×19px; at each scale, use the convolutional neural network CNN to extract features and obtain feature maps of different scales; fuse feature maps of different scales, and use weighted average or maximum fusion to form a multi-scale feature fusion map. Improve the robustness and comprehensiveness of feature expression.
[0067] S3: Construct a deep learning classification model, take the characteristics of the standard rice planting structure as input, and train the deep learning model once; specifically include the following steps: construct a deep learning classification model, using a ResNet model or a DenseNet model; take the characteristics of the standard rice planting structure as input, and train the deep learning model once; set the learning rate to 0.001, the batch size to 32, the number of iterations to 100 rounds, and use the cross-validation method to optimize the model.
[0068] S4: Use the trained deep learning model to classify the multi-scale feature fusion map, obtain the classification results of the rice planting structure, and calculate the area; specifically include the following steps: use the trained deep learning model to classify the multi-scale feature fusion map to obtain the classification layout of the rice planting structure; use morphological filtering and connected domain analysis to remove noise, fill holes, and optimize edge processing for the classification layout; count the number of pixels of the rice planting structure of each classification layout, and calculate the rice planting area of each classification layout according to the preset ratio classification. According to the experimental results and accuracy evaluation indicators, the parameters of the deep learning model are optimized to improve the extraction accuracy, and the performance of the model is evaluated by confusion matrix, ROC curve and other methods to ensure the stability and reliability of the model.
[0069] S5: In the deep learning classification model, input the standard feature parameters of the color, texture and shape framework of rice of different classifications for secondary training; specifically including the following steps: collect the standard feature parameters of the color, texture and shape framework of rice of different classifications from the relevant knowledge base to form a data set; the standard feature parameters of each category cover the rice structural variables of this category of rice at different growth stages and under different environmental conditions; in the deep learning classification model, select ResNet as the model architecture for rice feature extraction and classification, adjust the model and add convolutional layers; input the data set into the deep learning model for secondary training, and adopt cross-validation and early stopping strategies to avoid model overfitting. Color features: For rice of different classifications, the colors of its leaves, stems and panicles are different. The color features of rice, such as RGB values, HSV values, etc., are extracted by using the visible light and near-infrared band information in remote sensing images, and a color feature parameter library for rice of different classifications is constructed through statistical analysis. Texture features: The texture features of rice are mainly reflected in the arrangement of leaves, the thickness of stems and the density of panicles. The texture features of rice are extracted using texture description methods such as gray-level co-occurrence matrix (GLCM) and local binary pattern (LBP). Similarly, a texture feature parameter library for different classifications of rice is constructed. Shape features: The shape features of rice mainly include the shape of leaves, the curvature of stems and the shape of panicles. The shape features of rice are extracted using image processing techniques such as edge detection and contour extraction, and a shape feature parameter library for different classifications of rice is constructed.
[0070] S6: Use the trained deep learning model to compare the classified rice planting structures separately, obtain the qualified rate of each type of rice, and determine the abnormal structure area; specifically include the following steps: use the trained deep learning model to compare the classified rice planting structures separately, and use the color, texture and shape framework that all meet the standard feature parameters as the qualified standard to further refine each type of rice planting structure; determine the ratio of the qualified rice structure to the area of each type of rice planting structure layout, and determine its qualified rate; use the spatial information of the remote sensing image to mark the abnormal area on the image to form the abnormal structure area. Inputting the standard feature parameters of the color, texture and shape framework of different classifications of rice for secondary training is an effective method to improve the recognition accuracy of the deep learning classification model for rice planting structure. Through this method, the accurate extraction of rice planting area and real-time monitoring of growth status can be achieved, providing strong technical support for agricultural production. Definition of qualified rate: The qualified rate refers to the proportion of individuals in a certain type of rice that meet the normal growth standards (color, texture, shape, etc.). Normally grown rice is defined by setting a series of standard parameters. For each type of rice after classification, calculate the number of individuals and count the number of individuals that meet the normal growth standards. Then, divide the number of individuals that meet the standards by the total number of individuals to get the qualified rate of this type of rice. Abnormal structure refers to rice individuals or areas that differ greatly from the normal growth standards. These abnormalities include poor growth, pest infection, insufficient soil nutrients and other problems. Determine abnormal structure areas: By comparing the classified rice with the normal growth standards, identify those individuals or areas that do not meet the standards. Using the spatial information of remote sensing images, these abnormal individuals or areas are marked on the images to form abnormal structure areas. By analyzing the soil conditions and climate data in the abnormal areas, the causes of abnormal growth are explored. This helps to formulate targeted agricultural management measures to improve the yield and quality of rice.
[0071] S7: Construct the rice pests and diseases ontology knowledge base, convert it into standard structural parameters, and input it into the deep learning classification model for three trainings; specifically, the following steps are included: using external sources, expert libraries, text libraries and corresponding traditional knowledge bases as data sources, constructing the rice pests and diseases ontology knowledge base, and converting it into specific subdivided color, texture and shape framework structural parameters for different classifications of rice; inputting the above parameters of different classifications of rice into the deep learning classification model, corresponding to the abnormal structural areas of different classifications of rice, and using cross-validation for three trainings. The rice pests and diseases ontology knowledge base is a systematic and structured knowledge collection, which covers key information such as various types, symptoms, transmission routes, and prevention and control measures of rice pests and diseases. Collect a large amount of rice pests and diseases related information, including academic literature, agricultural manuals, expert experience, etc. Organize and analyze these materials, extract key information, and construct the knowledge base framework according to the ontology engineering method. During the construction process, define the classes, attributes, relationships, etc. of pests and diseases to ensure the integrity and accuracy of the knowledge base. Convert the information in the knowledge base into structured parameters. This is achieved by defining a series of feature vectors, labels, relationship matrices, etc.
[0072] S8: Classify the abnormal structure area using the trained deep learning model to obtain the abnormal type classification results of different abnormal structure rice in the abnormal structure area. Specifically including the following steps: use the forward propagation of the trained deep learning model to classify the abnormal structure area of each classified rice again; determine the types and distribution of different pests and diseases in each classified rice abnormal structure area. Output the final extraction result to the relevant expert website to obtain solutions for the abnormal area. According to the previous steps, use the deep learning model to classify the rice planting area and determine the abnormal structure area. Apply the trained deep learning model to the classification task of the abnormal structure area. Select a convolutional neural network (CNN) to process image data, or select a recurrent neural network (RNN) to process sequence data, and classify the rice in the abnormal structure area through the forward propagation of the model to obtain the abnormal type classification results of different abnormal structure rice. Analyze the classification results to determine the types and distribution of rice pests and diseases in the abnormal structure area. It is helpful to formulate targeted prevention and control measures to improve the yield and quality of rice.
[0073] The rice planting structure remote sensing extraction method based on deep learning obtains multi-phase satellite remote sensing image data of the rice area, obtains remote sensing images based on multi-scale image pyramids, and fuses them to obtain multi-scale feature fusion maps, builds a deep learning classification model, and performs one training based on the characteristics of the standard rice planting structure, thereby classifying the multi-scale feature fusion map, obtaining the classification results of the rice planting structure, and performing area measurement, thereby improving the accuracy and efficiency of rice planting area extraction, providing a more accurate scientific basis for agricultural management and decision-making, combining the powerful feature extraction ability of deep learning and the advantages of multi-scale fusion, and enhancing the model's adaptability to complex terrain and diversified farmland environments. And the present invention performs secondary training based on the classification results of the rice planting structure obtained from one training, using the standard feature parameters of the color, texture and shape framework of different classification rice, thereby comparing rice with different classification structures, based on the standard feature parameters as thresholds, obtaining the abnormal structure area of each type of rice structure, thereby facilitating the determination of the specific distribution area of the planting problem, and providing a basis for the overall rice quality assessment. At the same time, the present invention uses the rice pest and disease ontology knowledge base as the standard structure parameter based on the abnormal structure area obtained by the secondary training, and inputs the model for three trainings, thereby classifying the abnormal structure area, and obtaining the specific categories and distribution of pests and diseases of rice with different abnormal structures in the abnormal structure area, which is convenient for targeted remediation and solving the problem.
[0074] The above shows and describes the basic principles, main features and advantages of the present invention. It should be understood by those skilled in the art that the present invention is not limited to the above embodiments. The above embodiments and descriptions are only for explaining the principles of the present invention. Without departing from the spirit and scope of the present invention, the present invention may have various changes and improvements, which fall within the scope of the present invention. The scope of protection of the present invention is defined by the attached claims and their equivalents.
Claims
1. A rice planting structure remote sensing extraction method based on deep learning, characterized in that: The following steps are involved: S1: Obtain multi-temporal satellite remote sensing image data of rice areas, pre-process the data, synthesize bands, and calculate spectral indices; S2: Construct a multi-scale image pyramid, obtain remote sensing images at different scales, extract features at each scale, and fuse them to obtain a multi-scale feature fusion map; S3: Build a deep learning classification model, take the characteristics of the standard rice planting structure as input, and train the deep learning model once; S4: Use the trained deep learning model to classify the multi-scale feature fusion map, obtain the classification results of rice planting structure, and perform area measurement; S5: In the deep learning classification model, standard feature parameters of color, texture and shape frameworks for different classifications of rice are input for secondary training; S6: Use the trained deep learning model to compare the classified rice planting structures, obtain the qualified rate of each type of rice, and determine the abnormal structure area; S7: Construct a rice pest and disease ontology knowledge base, convert it into standard structure parameters, and input it into a deep learning classification model for three trainings; S8: Use the trained deep learning model to classify the abnormal structure area and obtain the abnormal type classification results of different abnormal structure rice in the abnormal structure area.
2. The method for remote sensing extraction of rice planting structure based on deep learning according to claim 1, characterized in that: The S1 comprises the following steps: Obtain multi-temporal and multi-spectral satellite remote sensing image data of rice areas, including visible light, near infrared and short-wave infrared bands; Preprocess the remote sensing image data, perform radiation calibration, atmospheric correction, geometric correction and noise removal in sequence; The preprocessed remote sensing image data was synthesized, the NDVI normalized difference vegetation index was calculated, and the characteristic information related to rice growth was extracted.
3. The method for remote sensing extraction of rice planting structure based on deep learning according to claim 1, characterized in that: The S2 comprises the following steps: Construct a multi-scale image pyramid to divide remote sensing image data into images of different scales, including 7px×7px, 13px×13px, and 19px×19px; At each scale, convolutional neural network (CNN) is used to extract features and obtain feature maps of different scales. Feature maps of different scales are fused by weighted average or maximum fusion to form a multi-scale feature fusion map.
4. The method for remote sensing extraction of rice planting structure based on deep learning according to claim 1, characterized in that: The S3 comprises the following steps: Build a deep learning classification model, using the ResNet model or the DenseNet model; The deep learning model is trained once using the features of the standard rice planting structure as input; The learning rate is set to 0.001, the batch size is set to 32, the number of iterations is set to 100, and the cross-validation method is used to optimize the model.
5. The method for remote sensing extraction of rice planting structure based on deep learning according to claim 1, characterized in that: The S4 comprises the following steps: The trained deep learning model is used to classify the multi-scale feature fusion map to obtain the classification layout of rice planting structure; Morphological filtering and connected domain analysis are used to remove noise, fill holes, and optimize edge processing of the classification layout; The number of pixels of the rice planting structure in each classification layout is counted, and the rice planting area of each classification layout is calculated according to the preset ratio classification.
6. The method for remote sensing extraction of rice planting structure based on deep learning according to claim 1, characterized in that: The S5 comprises the following steps: Collect standard feature parameters of color, texture and shape frameworks for different classifications of rice from relevant knowledge bases to form a data set; The standard characteristic parameters of each category cover the rice structural variables of this category of rice at different growth stages and under different environmental conditions; In the deep learning classification model, ResNet was selected as the model architecture for rice feature extraction and classification, and the model was adjusted to add convolutional layers; The data set is input into the deep learning model for secondary training, using cross-validation and early stopping strategies to avoid model overfitting.
7. The method for remote sensing extraction of rice planting structure based on deep learning according to claim 1, characterized in that: The S6 comprises the following steps: The trained deep learning model is used to compare the classified rice planting structures separately, and the rice planting structures of each category are further refined based on the color, texture and shape framework meeting the standard feature parameters; Determine the ratio of qualified rice structures to the rice area of each type of rice planting structure layout, and determine its qualified rate; By using the spatial information of remote sensing images, abnormal areas are marked on the images to form abnormal structural areas.
8. The method for remote sensing extraction of rice planting structure based on deep learning according to claim 1, characterized in that: The S7 comprises the following steps: Using external sources, expert databases, text databases and corresponding traditional knowledge bases as data sources, we build a rice pest and disease ontology knowledge base and convert it into specific color, texture and shape framework structure parameters for different classifications of rice. The parameters of the above-mentioned different classifications of rice were input into the deep learning classification model, and cross-validation was used to perform three trainings corresponding to the abnormal structural regions of the different classifications of rice.
9. The method for remote sensing extraction of rice planting structure based on deep learning according to claim 1, characterized in that: The S8 comprises the following steps: Using the forward propagation of the trained deep learning model, the abnormal structural area of each classified rice is reclassified; Determine the types and distribution of different pests and diseases in each classified rice abnormal structure zone.
10. The method for remote sensing extraction of rice planting structure based on deep learning according to claim 9, characterized in that: In S8, the final extraction result is output to the relevant expert website to obtain the solution for the abnormal area.