Crop disease and pest identification method based on agricultural production service platform

Abstract

The invention discloses a crop disease and pest identification method based on an agricultural production service platform. The method comprises the following steps: collecting crop data information; the crop data information comprises crop environment data, crop remote sensing image data and crop physical image data; performing feature extraction on the crop environment data, the crop remote sensing image data and the crop physical image data to obtain an environment feature vector, a remote sensing image feature vector and an image feature vector; performing feature splicing on the environment feature vector, the remote sensing image feature vector and the image feature vector to obtain a comprehensive feature vector; inputting the comprehensive feature vector into a disease and insect pest recognition model for model training to obtain a trained disease and insect pest recognition model; inputting to-be-detected crop data information into the trained pest and disease identification model, and outputting a pest and disease identification result of the to-be-detected crop. According to the invention, rapid and accurate identification and dynamic monitoring of plant diseases and insect pests can be realized.

Description

Technical Field

[0001] The present invention relates to the field of digital agriculture, and in particular to a method for identifying crop diseases and insect pests based on an agricultural production service platform. Background Art

[0002] In modern agricultural technology, the identification of pests and diseases has always been a challenging topic. Pest and disease identification is related to the healthy growth and stable yield of crops, and is a key link in ensuring food security and improving the quality of agricultural products. Traditional pest and disease identification methods mainly rely on manual observation and professional knowledge judgment, which is not only inefficient but also prone to errors.

[0003] With the development of image processing and deep learning technologies, image-based automatic pest and disease identification systems have gradually become a research hotspot. However, existing image processing and deep learning technologies still have the following limitations: Background interference: In complex field environments, background noise and irrelevant elements may interfere with the identification of pests and diseases; Insufficient feature extraction: Existing feature extraction methods cannot fully extract the key features of pests and diseases, affecting the accuracy of identification.

[0004] Therefore, in order to solve the above problems, a crop disease and pest identification method based on an agricultural production service platform is needed, which can realize rapid and accurate identification and dynamic monitoring of diseases and pests. Summary of the invention

[0005] In view of this, the purpose of the present invention is to overcome the defects in the prior art and provide a crop disease and pest identification method based on an agricultural production service platform, which can realize rapid and accurate identification and dynamic monitoring of diseases and pests.

[0006] The crop pest identification method based on the agricultural production service platform of the present invention comprises:

[0007] Collecting crop data information; the crop data information includes crop environmental data, crop remote sensing image data and crop physical image data;

[0008] Feature extraction is performed on crop environmental data, crop remote sensing image data and crop physical image data respectively to obtain environmental feature vectors, remote sensing image feature vectors and image feature vectors;

[0009] Perform feature concatenation on the environmental feature vector, the remote sensing image feature vector and the image feature vector to obtain a comprehensive feature vector;

[0010] The comprehensive feature vector is input into the pest and disease identification model to train the model, and a trained pest and disease identification model is obtained;

[0011] The crop data information to be tested is input into the trained pest and disease identification model, and the pest and disease identification results of the crops to be tested are output.

[0012] Furthermore, the crop environmental data includes temperature, humidity and soil pH value; the crop remote sensing image data includes crop growth status images and pest and disease occurrence area images; the crop physical image data includes crop leaf images and crop fruit images.

[0013] Furthermore, before feature extraction, the collected crop data information is preprocessed, specifically including:

[0014] For image data, remove noise, irrelevant background and unclear images; for environmental data, fill in missing values, remove outliers and perform standardization;

[0015] For image data, it is enhanced by rotation, scaling, cropping, and color transformation; for environmental data, it is enhanced by interpolation and time series transformation;

[0016] For image data, radiation correction, geometric correction, cropping and enhancement are performed.

[0017] Furthermore, feature extraction is performed on crop environmental data, including:

[0018] The crop environment data is converted into an embedding vector of a fixed size, and features are extracted from the embedding vector to obtain an environmental feature vector.

[0019] Furthermore, feature extraction is performed on crop remote sensing image data, including:

[0020] The disease-prominent area in the remote sensing image is extracted to obtain a saliency map, and features are extracted from the saliency map to obtain a remote sensing image feature vector; the remote sensing image feature vector includes color features, texture features, and shape features of the diseased spot.

[0021] Furthermore, feature extraction is performed on the crop physical image data, specifically including:

[0022] The image features in the physical image data are extracted through convolution, pooling, and full connection operations to obtain the image feature vector.

[0023] Furthermore, the comprehensive feature vector is input into the pest and disease identification model for model training, which specifically includes:

[0024] Set the learning rate, batch size, and number of iterations;

[0025] The comprehensive feature vector is used as input, and the features are first transformed nonlinearly through a fully connected layer, and then the features are weighted through an attention mechanism, and finally the final prediction results are output through another fully connected layer; the annotated pest and disease category is used as a label, a loss function is constructed, and the gap between the model prediction result and the true label is calculated to obtain the loss value; the gradient of the model parameters is calculated based on the loss value, and the model parameters are updated through back propagation;

[0026] If the model converges or reaches the preset number of training rounds, the model training ends.

[0027] Furthermore, the loss function Loss is constructed according to the following method:

[0028] Loss = α loss e +β·loss r +γ·loss i ;

[0029] Among them, α, β and γ are weight coefficients, and the sum of the three is 1; loss e Represents the mean square error loss corresponding to the environment feature vector; loss r Represents the cross entropy loss corresponding to the remote sensing image feature vector; loss i Represents the cross entropy loss corresponding to the image feature vector.

[0030] The beneficial effects of the present invention are as follows: the present invention discloses a method for identifying crop diseases and pests based on an agricultural production service platform, which highlights key areas of diseases and pests and reduces background interference through salient map extraction technology; improves the model's learning and identification capabilities for disease and pest characteristics through in-depth analysis of lesion characteristics; through a trained disease and pest identification model, it can monitor the occurrence and development of diseases and pests in real time, and provide real-time early warning for agricultural production; based on fused image, meteorological, soil and other multimodal data, it can not only improve the accuracy of disease and pest identification, but also enhance the model's adaptability to complex environmental changes, providing more reliable technical support for agricultural production. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] The present invention will be further described below in conjunction with the accompanying drawings and embodiments:

[0032] Figure 1 This is a schematic diagram of the process of the crop pest identification method of the present invention;

[0033] Figure 2 This is a schematic diagram of the overall implementation of crop pest and disease identification according to the present invention;

[0034] Figure 3 This is a schematic diagram of the pest and disease identification model training of the present invention. DETAILED DESCRIPTION

[0035] The present invention is further described below in conjunction with the accompanying drawings, as shown in the drawings:

[0036] This embodiment discloses a method for identifying crop diseases and insect pests based on an agricultural production service platform, comprising the following steps:

[0037] Collecting crop data information; the crop data information includes crop environmental data, crop remote sensing image data and crop physical image data;

[0038] Feature extraction is performed on crop environmental data, crop remote sensing image data and crop physical image data respectively to obtain environmental feature vectors, remote sensing image feature vectors and image feature vectors;

[0039] Perform feature concatenation on the environmental feature vector, the remote sensing image feature vector and the image feature vector to obtain a comprehensive feature vector;

[0040] The comprehensive feature vector is input into the pest and disease identification model to train the model, and a trained pest and disease identification model is obtained;

[0041] The crop data information to be tested is input into the trained pest and disease identification model, and the pest and disease identification results of the crops to be tested are output.

[0042] In this embodiment, the constructed agricultural production service platform is used to collect crop data information and analyze the collected crop data information to realize the identification of crop diseases and pests. Among them, the agricultural production service platform includes agricultural cloud (private cloud / public cloud), IoT perception work center, data intelligence work center, business collaboration work center, and intelligent applications.

[0043] The crop environmental data include temperature, humidity and soil pH value; the crop remote sensing image data include crop growth status images and pest and disease occurrence area images; the crop physical image data include crop leaf images and crop fruit images.

[0044] Among them, the corresponding crop environmental data can be collected by deploying various sensors, such as temperature, humidity, and soil pH sensors; remote sensing equipment can be used to obtain crop remote sensing image data; and image acquisition equipment can be used to obtain crop physical image data.

[0045] By setting the above three data types, a comprehensive monitoring of crop growth conditions is formed. Environmental data provides basic information on crop growth conditions, remote sensing image data reflects the overall growth status of crops and the distribution of pests and diseases from a macro perspective, and physical image data can show the specific conditions of individual crops (such as leaves and fruits) in detail. This comprehensive data collection can more accurately analyze the health status of crops.

[0046] In this embodiment, before feature extraction, the collected crop data information is preprocessed, specifically including:

[0047] For image data, remove noise, irrelevant background and unclear images; for environmental data, fill in missing values, remove outliers and perform standardization;

[0048] For image data, rotation, scaling, cropping, and color transformation are performed to increase data diversity; for environmental data, interpolation and time series transformation are performed to enhance data;

[0049] For image data, radiation correction, geometric correction, cropping and enhancement are performed.

[0050] In this embodiment, a deep neural network can be used to construct a sensor data analysis model, and time series analysis, principal component analysis (PCA) and other methods are used based on the sensor data analysis model to extract features of crop environmental data, specifically including:

[0051] The crop environmental data is converted into an embedding vector of a fixed size, and the embedded vector is subjected to feature extraction through operations such as a fully connected layer to obtain an environmental feature vector.

[0052] Convolutional neural network (CNN) or other deep learning models are used to build remote sensing image analysis models, and feature extraction of crop remote sensing image data is performed based on the remote sensing image analysis models, including:

[0053] The disease-prominent area in the remote sensing image is extracted to obtain a saliency map, and features are extracted from the saliency map to obtain a remote sensing image feature vector; the remote sensing image feature vector includes color features, texture features, and shape features of the diseased spot.

[0054] Convolutional neural networks (CNN), such as VGGNet and ResNet, are used to build image recognition models. Feature extraction of crop image data is performed based on the image recognition model, including:

[0055] The image features in the physical image data are extracted through convolution, pooling, and full connection operations to obtain the image feature vector.

[0056] Through the above feature extraction, the original and complex crop data information can be converted into concise and efficient feature representation. These feature vectors not only contain key information about the crop growth environment, overall growth status and individual details, but also facilitate subsequent further processing and identification, thereby improving the accuracy and efficiency of pest and disease identification.

[0057] In this embodiment, the above three feature vectors are spliced ​​in dimension to form a comprehensive feature vector, and the comprehensive feature vector is input into the pest and disease recognition model for model training, wherein the pest and disease recognition model can use a deep learning framework, such as TensorFlow or PyTorch, and select a suitable model architecture, such as MobileNetV2, ShuffleNet, etc. for construction, which will not be repeated here. The model training specifically includes:

[0058] Set the learning rate, batch size, and number of iterations; the learning rate is used to control the magnitude of model parameter updates; the batch size is used to control the amount of data used in each iteration; and the number of iterations is used to control the time for model training.

[0059] The comprehensive feature vector is taken as input, and the features are first transformed nonlinearly through a fully connected layer. Then the features are weighted through the attention mechanism, and finally the final prediction results are output through another fully connected layer. Among them, the labeled pest and disease categories are used as labels, and a loss function is constructed to calculate the gap between the model prediction results and the true labels to obtain the loss value. The gradient of the model parameters is calculated according to the loss value, and the model parameters are updated through back propagation. The optimizer, such as Adam, SGD, etc., can be used to update the model parameters.

[0060] If the model converges or reaches the preset number of training rounds, the model training ends. The goal of iterative training is to minimize the loss function, that is, to minimize the gap between the model prediction result and the true label.

[0061] In this embodiment, the loss function Loss is constructed according to the following method:

[0062] Loss = α loss e +β·loss r +γ·loss i ;

[0063] Among them, α, β and γ are weight coefficients, and the sum of the three is 1; loss e Represents the mean square error loss corresponding to the environment feature vector; loss r Represents the cross entropy loss corresponding to the remote sensing image feature vector; loss i Represents the cross entropy loss corresponding to the image feature vector.

[0064] Image feature vectors and remote sensing image feature vectors are usually classification tasks, in which case cross entropy loss is used, while environment feature vectors are usually regression tasks, in which case mean square error loss is used. By accepting model outputs and true labels from three different data sources, the respective losses are calculated separately.

[0065] The loss function constructed above can effectively combine environmental data, remote sensing images and physical image features, thereby further improving the accuracy and robustness of the recognition results.

[0066] Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solution of the present invention can be modified or replaced by equivalents without departing from the purpose and scope of the technical solution of the present invention, which should be included in the scope of the claims of the present invention.

Claims

1. A method for identifying crop diseases and insect pests based on an agricultural production service platform, characterized in that: include: Collecting crop data information; the crop data information includes crop environmental data, crop remote sensing image data and crop physical image data; Feature extraction is performed on crop environmental data, crop remote sensing image data and crop physical image data respectively to obtain environmental feature vectors, remote sensing image feature vectors and image feature vectors; Perform feature concatenation on the environmental feature vector, the remote sensing image feature vector and the image feature vector to obtain a comprehensive feature vector; The comprehensive feature vector is input into the pest and disease identification model to train the model, and a trained pest and disease identification model is obtained; The crop data information to be tested is input into the trained pest and disease identification model, and the pest and disease identification results of the crops to be tested are output.

2. The method for identifying crop diseases and insect pests based on an agricultural production service platform according to claim 1, characterized in that: The crop environmental data include temperature, humidity and soil pH value; the crop remote sensing image data include crop growth status images and pest and disease occurrence area images; the crop physical image data include crop leaf images and crop fruit images.

3. The method for identifying crop diseases and insect pests based on an agricultural production service platform according to claim 1, characterized in that: Before feature extraction, the collected crop data information is preprocessed, including: For image data, remove noise, irrelevant background and unclear images; for environmental data, fill in missing values, remove outliers and perform standardization; For image data, it is enhanced by rotation, scaling, cropping, and color transformation; for environmental data, it is enhanced by interpolation and time series transformation; For image data, radiation correction, geometric correction, cropping and enhancement are performed.

4. The method for identifying crop diseases and insect pests based on an agricultural production service platform according to claim 1, characterized in that: Feature extraction of crop environmental data, including: The crop environment data is converted into an embedding vector of a fixed size, and features are extracted from the embedding vector to obtain an environmental feature vector.

5. The method for identifying crop diseases and insect pests based on an agricultural production service platform according to claim 1, characterized in that: Feature extraction of crop remote sensing image data, including: The disease-prominent area in the remote sensing image is extracted to obtain a saliency map, and features are extracted from the saliency map to obtain a remote sensing image feature vector; the remote sensing image feature vector includes color features, texture features, and shape features of the diseased spot.

6. The method for identifying crop diseases and insect pests based on an agricultural production service platform according to claim 1, characterized in that: Feature extraction of crop image data includes: The image features in the physical image data are extracted through convolution, pooling, and full connection operations to obtain the image feature vector.

7. The method for identifying crop diseases and insect pests based on an agricultural production service platform according to claim 1, characterized in that: The comprehensive feature vector is input into the pest and disease identification model for model training, including: Set the learning rate, batch size, and number of iterations; The comprehensive feature vector is used as input, and the features are first transformed nonlinearly through a fully connected layer, and then the features are weighted through an attention mechanism, and finally the final prediction results are output through another fully connected layer; the annotated pest and disease category is used as a label, a loss function is constructed, and the gap between the model prediction result and the true label is calculated to obtain the loss value; the gradient of the model parameters is calculated based on the loss value, and the model parameters are updated through back propagation; If the model converges or reaches the preset number of training rounds, the model training ends.

8. The method for identifying crop diseases and insect pests based on an agricultural production service platform according to claim 7, characterized in that: The loss function Loss is constructed according to the following method: Loss=a·loss e +β·loss r +γ·loss i ; Among them, α, β and γ are weight coefficients, and the sum of the three is 1; loss e Represents the mean square error loss corresponding to the environment feature vector; loss r Represents the cross entropy loss corresponding to the remote sensing image feature vector; loss i Represents the cross entropy loss corresponding to the image feature vector.

Images