Identification method of citrus red spider pests based on deep learning

Through deep learning technology, combined with semantic segmentation and convolutional neural network, the problem of low recognition accuracy of crop pests is solved, and high-precision citrus red spider insect recognition is achieved, with the recognition accuracy reaching 98%, and training is reduced on small data sets Complexity.

CN112597907BActive Publication Date: 2024-11-22SICHUAN TECH & BUSINESS UNIV
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202011562695.4
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2020-12-25
Publication Date
2024-11-22
Estimated Expiration
2040-12-25

AI Technical Summary

Technical Problem

In the prior art, crop pest recognition depends on human eyes to be easily disturbed, the recognition accuracy is low, and the texture, shape and color characteristics of the pest image are easily affected by rotation, translation, and brightness, resulting in poor recognition effect.

Method used

Using the deep learning-based citrus spider pest recognition method, preprocessing and data label calibration is performed by collecting images, deep learning data sets are constructed, and trained in a convolutional neural network. Feature points are extracted using the Mask R-CNN model, combined with semantics Segmentation technology segments and analyzes images, and sets different analytical weights to identify different diseases.

Benefits of technology

High-precision citrus spider insect recognition is achieved, with an identification accuracy of 98%. Through transfer learning, training complexity and learning time are reduced on small data sets, and the recognition of different diseases is adapted to the recognition of different diseases.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN112597907B_ABST
    Figure CN112597907B_ABST
Patent Text Reader

Abstract

The present invention relates to the field of crop pest identification, and discloses a method for identifying citrus red spider pests based on deep learning, comprising: collecting citrus red spider pest images, and preprocessing the citrus red spider pest images; calibrating image data labels to produce a deep learning data set; training the deep learning data set in a convolutional neural network to obtain a trained citrus red spider pest identification model; inputting the image to be identified into the citrus red spider pest identification model to obtain the identification result of the citrus red spider pest identification model. The present invention solves the problem that the existing crop pest identification relies on human eye identification, which is easily disturbed and has low identification accuracy.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the field of crop disease and insect pest identification, and specifically refers to a method for identifying citrus red spider pests based on deep learning. Background Art

[0002] Citrus producing areas often have pests, mainly red spider mites. The daily red spider pest management in my country's citrus orchards mainly relies on the old management model, which is to fight insects when they are seen and prevent diseases when they are seen. It relies solely on herbicides to control weeds in orchards, which seriously destroys the ecological balance between species in citrus orchards. With the maturity and development of computer vision and image processing technology, it has become possible to use images to realize the automatic identification and diagnosis of citrus red spider pests. Intelligent identification of citrus red spider pests is a technical means to use imaging systems to complete processing and interpretation tasks through computers. It can realize the image segmentation, feature value extraction and automatic identification of citrus red spider pests.

[0003] At present, researchers at home and abroad are studying pest recognition methods by finding the texture, shape, color and other features of pest images to establish a feature database for recognition. However, these features are easily affected by rotation, translation and brightness, which restricts the recognition of pest images. Summary of the invention

[0004] Based on the above technical problems, the present invention provides a method for identifying citrus red spider pests based on deep learning, which solves the problem that the existing crop pest identification relies on human eye recognition, is susceptible to interference and has low recognition accuracy.

[0005] In order to solve the above technical problems, the technical solution adopted by the present invention is as follows:

[0006] A deep learning-based identification method for citrus red spider mite pests comprises: collecting citrus red spider mite pest images and preprocessing the citrus red spider mite pest images; calibrating data labels for image data to create a deep learning data set; training the deep learning data set in a convolutional neural network to obtain a trained citrus red spider mite pest identification model; inputting the image to be identified into the citrus red spider mite pest identification model to obtain the identification result of the citrus red spider mite pest identification model.

[0007] As a preferred method, the deep learning dataset is trained in a convolutional neural network to obtain a trained citrus red spider mite pest recognition model, including: constructing a Mask R-CNN model; extracting feature points of samples in the deep learning dataset; training the Mask R-CNN model through the feature points; and bringing the Mask R-CNN model into the convolutional neural network to obtain a citrus red spider mite pest recognition model.

[0008] As a preferred manner, the feature points include texture, shape, size and color of the citrus red spider in the citrus red spider image.

[0009] As a preferred method, training the Mask R-CNN model through feature points includes: dividing the citrus red spider pest image into a symptom-inducing area, a secondary continued analysis area and a wide collection area through semantic segmentation, and setting image analysis weights for the symptom-inducing area, the secondary continued analysis area and the wide collection area from high to low; based on the high analysis weight, decoding the symptom-inducing area to extract symptom picture features to identify the main cause of the disease; based on the medium analysis weight, searching the secondary continued analysis area for a large range of symptoms and marking the symptoms; based on the low analysis weight, searching the wide collection area for diffuse symptoms; combining the data of the symptom-inducing area, the secondary continued analysis area and the wide collection area to realize citrus red spider pest recognition.

[0010] As a preferred approach, the convolutional neural network uses a deep convolutional neural network model such as AlexNet, VGG-16 or GoogLeNet.

[0011] As a preferred approach, the deep learning dataset includes a training set, a validation set, and a test set.

[0012] As a preferred method, data label calibration includes marking the pests, crops, background and pest categories in the citrus red spider pest image.

[0013] As a preferred method, the preprocessing includes cropping and normalizing the blank part of the citrus red spider pest image.

[0014] Compared with the prior art, the present invention has the following beneficial effects:

[0015] (1) The present invention uses a deep learning training method to identify crop diseases. Deep learning can identify a variety of crop diseases and solve the problem of manual feature extraction. The use of transfer learning training methods can reduce training complexity and learning time on small data sets.

[0016] (2) The algorithm of the present invention is very complete, the recognition accuracy has reached 98%, and the recognition cycle is short. The MaskR-CNN image recognition technology used completes the semantic segmentation of the target individual. This algorithm is used to first analyze the image, accurately locate the subject range, and identify different symptoms by setting different weights for different ranges, thereby accurately judging the symptoms.

[0017] (3) The present invention uses relevant theories of deep learning, including three different deep learning models: deep belief network (DBN), recurrent neural network (RNN) and convolutional neural network (CNN), as well as some principles of CNN and its organizational structure: there are convolution layer (Convolution Layer), pooling layer (Pooling Layer), activation function and regularization technique Dropourt et al., and summarizes the training process including back propagation, gradient calculation, etc. This algorithm is suitable for the training method transfer learning of small data sets. The training process used in the present invention is combined with transfer learning, and then the model is tested with its own data. The performance of the three algorithms is compared using overall accuracy, average precision, average recall and F1 value. The results verify the feasibility of deep convolutional neural network for crop disease recognition, and the test accuracy of various diseases under different algorithms is measured. The algorithm with higher accuracy is selected based on performance indicators. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] The present application will be further described by way of exemplary embodiments, which will be described in detail with reference to the accompanying drawings, in which:

[0019] Figure 1 Flow chart of the identification method of citrus red spider pests based on deep learning.

[0020] Figure 2 Image of citrus spider mite infestation.

[0021] Figure 3 Schematic diagram of the recognition range of the Mask R-CNN model.

[0022] Figure 4 Continue parsing the area image for the secondary.

[0023] Figure 5 This is a schematic diagram of the analysis and identification of the secondary continued analysis area.

[0024] Figure 6 For wide collection of images

[0025] Figure 7 This is a schematic diagram for analyzing and identifying a wide collection area. DETAILED DESCRIPTION

[0026] In order to make the purpose, technical solution and advantages of the embodiments of the present disclosure clearer, the technical solution of the embodiments of the present disclosure will be clearly and completely described below in conjunction with the drawings of the embodiments of the present disclosure. Obviously, the described embodiments are part of the embodiments of the present disclosure, not all of the embodiments. Based on the described embodiments of the present disclosure, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present disclosure.

[0027] Figure 1 This is a flow chart of a method for identifying citrus red spider pests based on deep learning as shown in some embodiments of the present application. Figure 1 The present invention introduces the method for identifying citrus red spider pests based on deep learning. It should be noted that: Figure 1 This is only an example and does not limit the specific process of the deep learning-based method for identifying citrus red spider pests.

[0028] See also Figure 1 In this embodiment, a method for identifying citrus red spider mite pests based on deep learning includes: collecting citrus red spider mite pest images and preprocessing the citrus red spider mite pest images; calibrating image data labels to create a deep learning data set; training the deep learning data set in a convolutional neural network to obtain a trained citrus red spider mite pest recognition model; inputting the image to be recognized into the citrus red spider mite pest recognition model to obtain a recognition result of the citrus red spider mite pest recognition model.

[0029] In some embodiments, the deep learning dataset is trained in a convolutional neural network to obtain a trained citrus red spider mite pest recognition model, including: constructing a Mask R-CNN model; extracting feature points of samples in the deep learning dataset; training the Mask R-CNN model through the feature points; and bringing the Mask R-CNN model into the convolutional neural network to obtain a citrus red spider mite pest recognition model.

[0030] In this embodiment, Mask R-CNN is a network architecture based on Faster R-CNN, which mainly completes the semantic segmentation of the target individual. The idea of ​​the paper is mainly to expand the original Faster R-CNN, adding a branch to predict the existing target, and has a higher recognition rate for highly recognizable features.

[0031] In some embodiments, the feature points include texture, shape, size, and color of the citrus spider mite in the citrus spider mite image.

[0032] In some embodiments, training the Mask R-CNN model through feature points includes: dividing the citrus red spider pest image into a symptom-inducing area, a secondary continued analysis area, and a wide collection area through semantic segmentation, and setting image analysis weights for the symptom-inducing area, the secondary continued analysis area, and the wide collection area from high to low; based on the high analysis weight, decoding the symptom-inducing area to extract symptom picture features to identify the main cause of the disease; based on the medium analysis weight, searching the secondary continued analysis area for a large range of symptoms and marking the symptoms; based on the low analysis weight, searching the wide collection area for diffuse symptoms; combining the data of the symptom-inducing area, the secondary continued analysis area, and the wide collection area to realize citrus red spider pest identification.

[0033] In this embodiment, when using the Mask R-CNN model, the image is first analyzed to accurately locate the range of the subject, and different symptoms are identified by setting different weights for different ranges, thereby accurately determining the symptoms.

[0034] like Figure 2 and Figure 3 As shown, different image analysis weights are used in different regions. The high-resolution area can also be called the disease-inducing area. Based on the high-resolution weight, this area uses a higher resolution rate to decode and extract more disease image features to better identify the main cause of the disease. In the secondary continued analysis area, based on the medium-resolution weight, medium-resolution is used to search for diseases in a large range and mark them out. In the wide collection area, based on the low-resolution weight, the wide collection area simply collects diffuse symptoms.

[0035] Finally, the data from the disease-inducing area, secondary continued analysis area and wide collection area are combined to realize the identification of citrus red spider pests.

[0036] For specific large-scale identification such as Figure 4 and Figure 5 As shown, in the secondary continued analysis area, medium resolution is used to search for symptoms in a large range and mark them. If further identification of small-scale and multiple symptoms is required, the wide collection area can simply collect the diffuse symptoms and combine them with the data from the previous two areas for analysis to achieve accurate identification results, such as Figure 6 Figure 7 shown.

[0037] In some embodiments, the convolutional neural network adopts a deep convolutional neural network model AlexNet, VGG-16 or GoogLeNet.

[0038] In this embodiment, on the basis of convolutional neural network, three deep convolutional neural network models, AlexNet, VGG-16 and GoogLeNet, are used, and they are combined with transfer learning to analyze the recognition effects of different iterations and different algorithms. The results of various disease recognition rates, average recognition accuracy, recall rate, recall rate and F1 value show that the VGG-16 model has better effect, and the average recognition rate of diseases reaches 98.92%. It shows that the use of deep learning in crop disease recognition can identify a variety of crop diseases and solve the problem of artificial feature extraction. The training method of transfer learning can reduce the training complexity and learning time on small data sets.

[0039] In some embodiments, the deep learning dataset includes a training set, a validation set, and a test set.

[0040] In some embodiments, data labeling includes marking the pests, crops, background, and pest categories in the citrus red spider pest image.

[0041] In some embodiments, the preprocessing includes cropping and normalizing the blank portion of the citrus red spider pest image.

[0042] The above is an embodiment of the present invention. The above embodiments and the specific parameters in the embodiments are only for the purpose of clearly describing the verification process of the invention, and are not intended to limit the patent protection scope of the present invention. The patent protection scope of the present invention shall still be subject to its claims, and any equivalent structural changes made by using the contents of the description and drawings of the present invention shall be included in the protection scope of the present invention.

Claims

1. A method for identifying citrus red spider pests based on deep learning, characterized in that: include: Collecting citrus red spider mite pest images and preprocessing the citrus red spider mite pest images; Performing data label calibration on the image data to produce a deep learning data set; The deep learning data set is trained in a convolutional neural network to obtain a trained citrus red spider pest recognition model; the image to be recognized is input into the citrus red spider pest recognition model to obtain a recognition result of the citrus red spider pest recognition model; The step of training the deep learning dataset in a convolutional neural network to obtain a trained citrus red spider mite pest recognition model includes: constructing a Mask R-CNN model; extracting feature points of samples in the deep learning dataset; training the Mask R-CNN model using the feature points; and bringing the Mask R-CNN model into the convolutional neural network to obtain the citrus red spider mite pest recognition model. The training of the Mask R-CNN model through the feature points includes: dividing the citrus red spider pest image into a symptom-inducing area, a secondary continued analysis area and a wide collection area through semantic segmentation, and setting image analysis weights for the symptom-inducing area, the secondary continued analysis area and the wide collection area from high to low; based on the high analysis weight, decoding the symptom-inducing area to extract symptom picture features to identify the main cause of the disease; based on the medium analysis weight, searching the secondary continued analysis area for a large range of symptoms and marking the symptoms; based on the low analysis weight, searching the wide collection area for diffuse symptoms; combining the data of the symptom-inducing area, the secondary continued analysis area and the wide collection area to realize citrus red spider pest identification.

2. The method for identifying citrus red spider pests based on deep learning according to claim 1, characterized in that :The feature points include the texture, shape, size and color of the citrus red spider in the citrus red spider image.

3. The method for identifying citrus red spider pests based on deep learning according to claim 1, characterized in that: The convolutional neural network adopts a deep convolutional neural network model AlexNet, VGG-16 or GoogLeNet.

4. The method for identifying citrus red spider pests based on deep learning according to claim 1, characterized in that: The deep learning dataset includes a training set, a validation set, and a test set.

5. The method for identifying citrus red spider pests based on deep learning according to claim 1, characterized in that: The data label calibration includes marking the pests, crops, background and pest categories in the citrus red spider pest image.

6. The method for identifying citrus red spider pests based on deep learning according to claim 1, characterized in that: The preprocessing includes cutting and normalizing the blank part of the citrus red spider pest image.