Gramineae crop disease detection method based on deep ensemble learning
Through a method based on deep ensemble learning, multiple residual neural network detection models are established, and data augmentation and weighted averaging technology are used to solve the problem of insufficient detection accuracy and generalization ability of grass crop diseases in the existing technology, achieving higher detection accuracy and adaptability.
Patent Information
- Application Number
- CN202411503006.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-10-25
- Publication Date
- 2025-03-04
AI Technical Summary
The existing methods for detecting diseases in grass family crops have significant shortcomings in detection accuracy and generalization capabilities, especially in complex agricultural environments, which are difficult to effectively identify diverse disease types.
Using a method based on deep ensemble learning, the robustness and generalization ability of the detection model are improved by establishing multiple residual neural network detection models, and through technical means such as data augmentation, pixel value normalization and weighted average.
It significantly improves the accuracy and generalization ability of grass family crop disease detection, can better identify diseases of different types of crops at different growth stages and under various environmental conditions, and is suitable for actual agricultural scenarios.
Smart Images
Figure CN119559423A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of agricultural biotechnology, and in particular to a gramineous crop disease detection method based on deep ensemble learning. Background Art
[0002] Currently, disease detection methods for gramineous crops mainly rely on traditional machine learning techniques and manually designed feature extraction methods. These methods may perform well under certain conditions, but in actual agricultural environments, due to the complex and changeable environment, the wide variety of gramineous crops and the variety of disease types, traditional methods have significant deficiencies in detection accuracy and generalization ability. Traditional feature extraction techniques cannot effectively cope with these challenges, and a single model is prone to overfitting, resulting in a significant decrease in performance on new data. Therefore, there is an urgent need for a solution that can improve the robustness and accuracy of disease detection to better serve the disease management of gramineous crops. Summary of the invention
[0003] In view of the above-mentioned deficiencies in the prior art, the present invention provides a method for detecting diseases of gramineous crops based on deep ensemble learning, which solves the problem that the existing disease detection methods have significant deficiencies in detection accuracy and generalization ability.
[0004] In order to achieve the above-mentioned object of the invention, the technical solution adopted by the present invention is:
[0005] A method for detecting diseases of gramineous crops based on deep ensemble learning is provided, which comprises:
[0006] Step 1: Data collection: Collect grass crop image datasets;
[0007] Step 2: Data preprocessing: preprocess the grass crop image dataset to obtain a training set;
[0008] Step 3: Establish a model: establish multiple residual neural network detection models and train the multiple detection models;
[0009] Step 4: Model evaluation and optimization: After completing the training of the detection model, evaluate the performance of the detection model through cross-validation and using an independent test set. Based on the evaluation results, optimize the detection model.
[0010] Step 5: Deploy the detection model to the actual production environment to perform real-time disease detection on grass crops.
[0011] Furthermore, in step 1, the grass crop image dataset includes healthy crop image data samples and various disease image data samples of different types of grass crops at different growth stages and under various environmental conditions. These datasets cover different types of grass crops and their disease manifestations at different growth stages, and cover disease image data under various environmental conditions. When training the detection model, the grass crop image dataset is used to enhance the generalization ability and robustness of the detection model.
[0012] Furthermore, in step 2, the original grass crop image dataset is first expanded using data augmentation technology to generate a diversified dataset to further improve the robustness of the detection model, and then the images in the grass crop image dataset are normalized to a specific range to ensure data consistency and model stability. Finally, the images are adjusted to the input size required by the model to obtain a training set.
[0013] Furthermore, data enhancement techniques include methods such as image rotation, flipping, scaling, shearing, and color jittering.
[0014] Furthermore, in step 3, the last fully connected layer of the residual neural network is replaced with a fully connected layer suitable for the binary classification task of gramineous crop diseases to establish a detection model; preferably, the detection model adds one or more fully connected layers and Dropout layers to improve the generalization ability of the detection model by randomly ignoring some neurons.
[0015] The method for training the detection model is as follows: the detection model includes multiple independent base learners, each base learner is a residual network model, the training set in step 2 is divided into multiple different and intersecting subsets, and an independent base learner is trained using each subset to obtain multiple diversified base learners; so that the detection model can better summarize the disease characteristics of grass crops and improve the overall detection performance.
[0016] Multiple diverse base learners predict the input image and output multiple prediction results. The multiple prediction results are summarized and compared with the validation set to determine the weighted value of the base learner corresponding to each prediction result. The final prediction result is obtained by weighted averaging. The weighted averaging method can reduce the uncertainty caused by the poor performance of individual base learners, reduce model variance, and effectively improve the reliability of prediction results, thereby improving the accuracy of disease detection.
[0017] Furthermore, methods for optimizing the detection model include changing the learning rate, number, and data enhancement techniques of the base learners to ensure that the detection model maintains efficient disease detection performance in practical applications. Through iterative optimization, the detection accuracy and robustness of the detection model are continuously improved to adapt it to different agricultural scenarios and crop disease types.
[0018] Furthermore, in step 5, the detection model is regularly adjusted using the newly collected grass crop image dataset to ensure the long-term stability and effectiveness of the detection model; the adjustment method is: multiple diversified base learners predict the new input image and output multiple prediction results, the multiple prediction results are aggregated and compared with the verification set, the weighted value of each prediction result is determined, and the final prediction result is obtained through the weighted average method to complete the adjustment of the detection model.
[0019] The beneficial effects of the present invention are: a method for detecting diseases of gramineous crops based on deep ensemble learning in this scheme uses a residual neural network to establish a detection model and train the detection model. The detection model has multiple independent base learners, and each base learner can output a prediction result for the input image, which significantly improves the accuracy and generalization ability of disease detection, and is particularly suitable for disease identification of gramineous crops. BRIEF DESCRIPTION OF THE DRAWINGS
[0020] Figure 1 The figure is a block diagram of a grass crop disease detection method based on deep ensemble learning.
[0021] Figure 2 Schematic diagram of the principle of the detection model training method. DETAILED DESCRIPTION
[0022] The specific implementation modes of the present invention are described below so that those skilled in the art can understand the present invention. However, it should be clear that the present invention is not limited to the scope of the specific implementation modes. For those of ordinary skill in the art, as long as various changes are within the spirit and scope of the present invention as defined and determined by the attached claims, these changes are obvious, and all inventions and creations utilizing the concept of the present invention are protected.
[0023] like Figure 1 As shown, a method for detecting diseases of gramineous crops based on deep ensemble learning comprises:
[0024] Step 1. Data collection: Collect grass crop image datasets; specifically, the grass crop image datasets include healthy crop image data samples and various disease image data samples of different types of grass crops at different growth stages and under various environmental conditions. These datasets cover different types of grass crops and their disease manifestations at different growth stages, and cover disease image data under various environmental conditions. When training the detection model, the grass crop image dataset is used to enhance the generalization ability and robustness of the detection model.
[0025] Step 2, data preprocessing: preprocess the grass crop image dataset to obtain a training set; the preprocessing method is: first use data augmentation technology to expand the original grass crop image dataset, data augmentation technology includes image rotation, flipping, scaling, shearing and color jittering and other methods to generate a diversified dataset, further improve the robustness of the detection model, and then perform pixel value normalization on the images in the grass crop image dataset, standardize the pixel values to a specific range to ensure data consistency and model stability, and finally adjust the image to the input size required by the model to obtain a training set.
[0026] Step 3, establish a model: establish multiple residual neural network detection models, and train multiple detection models; the specific method is: replace the last fully connected layer of the residual neural network with a fully connected layer suitable for the binary classification task of grass crop diseases to establish a detection model; preferably, the detection model adds one or more fully connected layers and Dropout layers to improve the generalization ability of the detection model by randomly ignoring some neurons.
[0027] The method for training the detection model is as follows: the detection model includes multiple independent base learners, each base learner is a residual network model, the training set in step 2 is divided into multiple different and intersecting subsets, and an independent base learner is trained using each subset to obtain multiple diversified base learners; so that the detection model can better summarize the disease characteristics of grass crops and improve the overall detection performance.
[0028] Multiple diverse base learners predict the input image and output multiple prediction results. The multiple prediction results are summarized and compared with the validation set to determine the weighted value of the base learner corresponding to each prediction result. The final prediction result is obtained by weighted averaging. The weighted averaging method can reduce the uncertainty caused by the poor performance of individual base learners, reduce model variance, and effectively improve the reliability of prediction results, thereby improving the accuracy of disease detection.
[0029] Step 4: Model evaluation and optimization: After completing the training of the detection model, evaluate the performance of the detection model through cross-validation and using an independent test set. Based on the evaluation results, optimize the detection model. Methods for optimizing the detection model include changing the learning rate, number, and data enhancement technology of the base learner to ensure that the detection model maintains efficient disease detection performance in practical applications. Through iterative optimization, the detection accuracy and robustness of the detection model are continuously improved to adapt it to different agricultural scenarios and crop disease types.
[0030] Step 5: Deploy the detection model to the actual production environment to perform real-time disease detection on gramineous crops. Regularly use newly collected gramineous crop image datasets to adjust the detection model to ensure the long-term stability and effectiveness of the detection model; the adjustment method is: multiple diversified base learners predict the new input image and output multiple prediction results, summarize the multiple prediction results and compare them with the validation set, determine the weighted value of each prediction result, and obtain the final prediction result through the weighted average method to complete the adjustment of the detection model.
[0031] In summary, this scheme uses a deep ensemble learning-based grass crop disease detection method to establish a detection model using a residual neural network and train the detection model. The detection model has multiple independent base learners, and each base learner can output a prediction result for the input image, which significantly improves the accuracy and generalization ability of disease detection, and is especially suitable for disease identification of grass crops.
Claims
1. A method for detecting diseases of gramineous crops based on deep ensemble learning, characterized in that: include Step 1: Data collection: Collect grass crop image datasets; Step 2: Data preprocessing: preprocess the grass crop image dataset to obtain a training set; Step 3: Establish a model: establish multiple residual neural network detection models and train the multiple detection models; Step 4: Model evaluation and optimization: After completing the training of the detection model, evaluate the performance of the detection model through cross-validation and using an independent test set. Based on the evaluation results, optimize the detection model. Step 5: Deploy the detection model to the actual production environment to perform real-time disease detection on grass crops.
2. The method for detecting diseases of gramineous crops based on deep ensemble learning according to claim 1, characterized in that: In step 1, the grass crop image dataset includes healthy crop image data samples and multiple disease image data samples of different types of grass crops at different growth stages and under multiple environmental conditions.
3. The method for detecting diseases of gramineous crops based on deep ensemble learning according to claim 1, characterized in that: In step 2, the original grass crop image dataset is first expanded using data augmentation technology, and then the images in the grass crop image dataset are normalized to a specific range of pixel values. Finally, the images are adjusted to the input size required by the model to obtain a training set.
4. The method for detecting gramineous crop diseases based on deep ensemble learning according to claim 3, characterized in that: Data augmentation techniques include image rotation, flipping, scaling, cropping, and color jittering.
5. The method for detecting gramineous crop diseases based on deep ensemble learning according to claim 4, characterized in that: In step 3, the last fully connected layer of the residual neural network is replaced with a fully connected layer suitable for the binary classification task of gramineous crop diseases to establish a detection model; The detection model training method is as follows: the detection model includes multiple independent base learners, each base learner is a residual network model, the training set in step 2 is divided into multiple different and intersecting subsets, and an independent base learner is trained using each subset to obtain multiple diversified base learners; Multiple diverse base learners predict the input image and output multiple prediction results. The multiple prediction results are summarized and compared with the validation set to determine the weighted value of the base learner corresponding to each prediction result. The final prediction result is obtained by weighted averaging.
6. The method for detecting gramineous crop diseases based on deep ensemble learning according to claim 5, characterized in that: In step 4, the method of optimizing the detection model includes changing the learning rate, number and data augmentation techniques of the base learners.
7. The method for detecting gramineous crop diseases based on deep ensemble learning according to claim 6, characterized in that: In step 5, the detection model is regularly adjusted using the newly collected grass crop image dataset; the adjustment method is: multiple diversified base learners predict the new input image and output multiple prediction results, the multiple prediction results are summarized and compared with the verification set, the weighted value of each prediction result is determined, and the final prediction result is obtained through the weighted average method to complete the adjustment of the detection model.