Litchi anthracnose disease resistance grade rapid identification system and method based on image identification

Abstract

The invention discloses a litchi anthracnose disease resistance grade rapid identification system and method based on image recognition, and the system comprises an image collection module which is used for collecting and obtaining an image covering the whole litchi leaf or fruit surface; the image processing module is used for preprocessing the image and then extracting features to obtain a scab image; the model construction module is used for constructing an initial deep convolutional neural network model, inputting the disease spot image into the initial deep convolutional neural network model for training, and obtaining a target deep convolutional neural network model; and the disease resistance level evaluation module is used for carrying out evaluation classification on the to-be-identified images of the surfaces of the litchi leaves or fruits in different stages and different types based on the target deep convolutional neural network model, and carrying out division to obtain a disease resistance level classification result. According to the invention, peasants and agricultural technicians can be helped to quickly and accurately identify diseases in practical application, and the control efficiency is improved.

Description

Technical Field

[0001] The invention belongs to the field of plant pathology, and in particular relates to a system and method for quickly identifying the disease resistance level of litchi anthracnose based on image recognition. Background Art

[0002] Litchi (Litchi chinensis Sonn.) is an important subtropical fruit, which is deeply loved by consumers for its delicious, juicy and unique taste. However, in recent years, with the promotion of intensive agricultural cultivation technology and the increase in planting density, coupled with insufficient management experience and complex and changeable climatic conditions, the litchi industry faces severe disease challenges. Among them, litchi anthracnose (Colletotrichum gloeosporioides) is one of the most destructive fungal diseases. It not only harms the leaves, flower spikes, branches and fruits of litchi in the field, causing flowers and fruits to fall off, but also causes browning and rot of fruits after harvest, seriously damaging the commercial value and storage performance of the fruits. At present, the prevention and control of anthracnose mainly relies on chemical agents, such as ergosterol biosynthesis inhibitors and dithiocarbamates, but the long-term use of these drugs will not only gradually weaken the effect, but also induce the enhancement of pathogen resistance, and have a negative impact on the ecological environment.

[0003] Faced with the threat of anthracnose, screening and breeding of disease-resistant varieties have become the key to ensuring the sustainable development of the litchi industry. However, in the current litchi planting industry, there is a lack of highly resistant varieties that can effectively resist anthracnose. Although in vitro inoculation and cluster analysis methods have been widely used in plant disease resistance evaluation and have achieved certain results, there are still deficiencies in the screening and evaluation of litchi anthracnose-resistant varieties. This is mainly reflected in the following aspects:

[0004] 1. Lack of a systematic resistance evaluation system: Current resistance evaluation mostly relies on traditional in vitro inoculation experiments, which is time-consuming, costly, and difficult to achieve large-scale variety screening.

[0005] 2. Manual evaluation is highly subjective: In traditional methods, the evaluation of lesion area and disease index often relies on manual observation, which is not only inefficient but also easily affected by personal experience and technical level, resulting in limited repeatability and accuracy of the evaluation results.

[0006] 3. Lack of real-time monitoring methods: Existing technologies make it difficult to achieve real-time monitoring and early warning of litchi anthracnose resistance, thus missing the best time for prevention and control.

[0007] 4. Lack of adaptability and accuracy: The existing resistance evaluation system may not be able to fully reflect the actual disease resistance performance under complex field environments, limiting the application effect of the selected varieties in actual production.

[0008] In recent years, with the rapid development of artificial intelligence and machine learning technologies, image recognition technology has shown great application potential in the field of plant disease diagnosis. Image recognition technology can quickly and accurately identify the characteristics of lesions on plant leaves and fruits, thus providing a new solution for early diagnosis and resistance assessment of plant diseases. Compared with traditional manual evaluation methods, image recognition technology has the following advantages:

[0009] 1. Efficiency: Image recognition technology can quickly process large amounts of data, greatly improving evaluation efficiency and reducing labor costs.

[0010] 2. Objectivity: Algorithm-based automatic identification reduces interference from human factors and improves the objectivity and consistency of evaluation results.

[0011] 3. Real-time: Real-time image processing capabilities make disease monitoring more timely and help with early detection and intervention.

[0012] 4. Scalability: Image recognition technology can be easily integrated into smart devices, such as smartphones, drones, etc., facilitating large-scale applications under field conditions.

[0013] In summary, the current screening and evaluation methods for litchi anthracnose-resistant varieties have obvious limitations, and image recognition technology provides a new way to overcome these limitations. Therefore, it is of great significance to use image recognition technology to build an efficient, accurate, and real-time method for rapid identification of litchi anthracnose resistance levels to fill the gap in existing technologies and contribute to the healthy and sustainable development of the litchi industry. Summary of the invention

[0014] In view of the problems of low efficiency, high cost and strong subjectivity in the current identification of the disease resistance level of litchi anthracnose, the present invention aims to develop a fast, accurate and low-cost method for identifying the disease resistance level of litchi anthracnose based on image recognition technology. Through intelligent image processing and analysis, this method can automatically identify and quantify the characteristics of anthracnose spots on litchi leaves or fruits, thereby quickly evaluating the disease resistance level of different litchi varieties, and providing strong technical support for the screening, cultivation and prevention and control of disease-resistant litchi varieties. The goal of the present invention is to greatly improve the identification efficiency and reduce manual errors on the basis of ensuring the accuracy and reliability of the identification results, so as to achieve real-time monitoring and early warning of the disease resistance of litchi anthracnose, and ultimately promote the sustainable development of the litchi industry and the improvement of economic benefits.

[0015] To achieve the above-mentioned purpose, the present invention provides a system and method for quickly identifying the disease resistance level of litchi anthracnose based on image recognition. Among them, a system for quickly identifying the disease resistance level of litchi anthracnose based on image recognition comprises:

[0016] An image acquisition module, used to acquire images covering the entire surface of litchi leaves or fruits;

[0017] An image processing module, connected to the image acquisition module, is used to perform feature extraction on the image after preprocessing to obtain a lesion image;

[0018] A model building module, connected to the image processing module, for building an initial deep convolutional neural network model, inputting the lesion image into the initial deep convolutional neural network model for training, and obtaining a target deep convolutional neural network model;

[0019] The disease resistance grade evaluation module is connected to the model building module and is used to evaluate and classify the images to be identified of litchi leaves or fruit surfaces at different stages and of different types based on the target deep convolutional neural network model, and obtain disease resistance grade classification results.

[0020] Preferably, the image processing module includes a preprocessing unit and a feature extraction unit;

[0021] The preprocessing unit is used to perform image denoising, color correction, brightness and contrast adjustment on the collected image, eliminate the influence of light and shadow on the image quality, and obtain a preprocessed image;

[0022] The feature extraction unit is used to extract features from the preprocessed image to obtain a lesion feature image.

[0023] Preferably, the preprocessing unit includes a format conversion unit, a cropping and scaling unit, an image enhancement unit, and an image denoising unit;

[0024] The format conversion unit is used to convert the collected images into a unified format;

[0025] The cropping and scaling unit is used to crop the image by edge detection or threshold segmentation, remove irrelevant background, separate the leaf or fruit area from the background, obtain the diseased spot area, and scale the image of the diseased spot area to a uniform size;

[0026] The image enhancement unit is used to apply data enhancement technology to perform rotation, flipping, color correction, brightness and contrast adjustment;

[0027] The image denoising unit is used to remove image noise through Gaussian filtering or median filtering.

[0028] Preferably, the feature extraction unit includes a color feature extraction unit, a texture feature extraction unit, a shape feature extraction unit, and a deep learning feature extraction unit;

[0029] The feature extraction unit is used to extract feature values ​​in RGB and HSV color spaces;

[0030] The texture feature extraction unit is used to extract texture features using a gray level co-occurrence matrix or a local binary pattern method;

[0031] The shape feature extraction unit is used to analyze the shape feature, size feature and edge feature of the lesion;

[0032] The deep learning feature extraction unit is used to automatically extract high-dimensional features through a convolutional neural network.

[0033] Preferably, the target deep convolutional neural network model is an improved ShuffleNet V2 model, which adopts a hybrid hole convolution and a lightweight channel attention module to enhance the feature extraction capability, uses an Adam optimizer, sets the initial learning rate to 0.001, and adopts a learning rate decay strategy, and adopts a cross entropy loss function as the loss function.

[0034] Preferably, the disease resistance grade assessment module includes a lesion quantification analysis unit, a disease resistance grade standard establishment unit, a cluster analysis unit, and a grade assessment unit;

[0035] The lesion quantitative analysis unit is used to identify and locate lesions in the image based on the target deep convolutional neural network model, obtain characteristic information of the lesions, and provide a quantitative evaluation index for disease resistance evaluation based on the characteristic information;

[0036] The disease resistance level standard establishment unit is used to combine historical data and expert knowledge to set the threshold range of each disease resistance level, and divide the varieties into four disease resistance levels: high resistance, medium resistance, medium sensitivity, and high sensitivity. Each disease resistance level corresponds to a certain range of lesion area and lesion severity.

[0037] The cluster analysis unit is used to analyze the quantitative data of lesions of different litchi varieties using a clustering algorithm on the lesion images, and to classify the varieties into different disease resistance levels by using the intra-group linkage and squared Euclidean distance method according to the lesion area and lesion severity;

[0038] The grade evaluation unit is used to establish a decision tree model or a support vector machine model using a machine learning algorithm, input a quantitative evaluation index, and output a corresponding disease resistance grade.

[0039] Preferably, the characteristic information of the lesions includes but is not limited to the area, number, distribution, shape, position and size of the lesions.

[0040] Preferably, the quantitative evaluation indicators include but are not limited to accuracy, recall rate, and F1 score.

[0041] The present invention also provides a method for quickly identifying the disease resistance level of litchi anthracnose based on image recognition, comprising the following steps:

[0042] Acquiring an image covering the entire surface of a litchi leaf or fruit, preprocessing the image, and then performing feature extraction to obtain a disease spot image;

[0043] Constructing an initial deep convolutional neural network model, inputting the lesion image into the initial deep convolutional neural network model for training, and obtaining a target deep convolutional neural network model;

[0044] Based on the target deep convolutional neural network model, the images to be identified on the surfaces of litchi leaves or fruits at different stages and of different types are evaluated and classified to obtain disease resistance grade classification results.

[0045] Compared with the prior art, the present invention has the following advantages and technical effects:

[0046] The present invention uses deep learning image recognition technology and cluster analysis to integrate the advantages of computer vision and machine learning to build an automated and intelligent system for rapid identification of litchi anthracnose resistance. The system first uses high-precision image acquisition equipment to capture high-definition images of litchi leaves or fruits, and then uses image preprocessing algorithms to remove interference and enhance the characteristics of lesions. Subsequently, a deep convolutional neural network (CNN) model analyzes the image, automatically identifies and locates lesions, and accurately extracts key information such as the morphology, size, and position of the lesions. On this basis, the system uses cluster analysis methods to classify the disease resistance of different litchi varieties according to the quantitative data of lesions, and finally achieves a rapid and accurate assessment of litchi anthracnose resistance.

[0047] The present invention organically combines the deep learning model with the clustering algorithm, which not only greatly improves the accuracy of lesion identification and the reliability of disease resistance level assessment, but also greatly improves the assessment efficiency and reduces manual participation and subjective bias. The technical solution of the present invention successfully solves the technical problems of low efficiency, high cost and large subjective judgment errors in traditional anthracnose resistance identification, provides a powerful tool for the screening of lychee disease-resistant varieties and the formulation of disease prevention and control strategies, promotes the sustainable development of the lychee industry, protects the economic interests of lychee growers, and also provides a reference model for the intelligent diagnosis of other crop diseases. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] The drawings constituting a part of the present application are used to provide a further understanding of the present application. The illustrative embodiments and descriptions of the present application are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings:

[0049] Figure 1 A schematic diagram of the system structure of an embodiment of the present invention;

[0050] Figure 2This is a comparison diagram of the lesion areas of different varieties of litchi leaves inoculated with anthracnose according to an embodiment of the present invention;

[0051] Figure 3 A cluster analysis diagram of the disease resistance of leaves of different litchi varieties to anthracnose bacteria according to an embodiment of the present invention;

[0052] Figure 4 This is a schematic diagram of the disease conditions of litchi leaves with different resistance levels according to an embodiment of the present invention;

[0053] Figure 5 This is a diagram showing the disease conditions of different varieties of litchi fruits after inoculation according to an embodiment of the present invention;

[0054] Figure 6 A cluster analysis diagram of the disease resistance of different litchi varieties to anthracnose bacteria according to an embodiment of the present invention;

[0055] Figure 7 This is a diagram showing the disease conditions of litchi fruits with different resistance levels according to an embodiment of the present invention. DETAILED DESCRIPTION

[0056] It should be noted that, in the absence of conflict, the embodiments and features in the embodiments of the present application can be combined with each other. The present application will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0057] It should be noted that the steps shown in the flowcharts of the accompanying drawings can be executed in a computer system such as a set of computer executable instructions, and that, although a logical order is shown in the flowcharts, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0058] Embodiment 1

[0059] like Figure 1 As shown, this embodiment provides a system for rapid identification of litchi anthracnose resistance grade based on image recognition, comprising:

[0060] An image acquisition module is used to acquire images covering the entire surface of litchi leaves or fruits, and obtain multiple litchi images including healthy leaves and fruits and anthracnose spots of varying degrees;

[0061] The image processing module is connected to the image acquisition module and is used to perform feature extraction on the image after preprocessing to obtain the lesion image;

[0062] A model building module, connected to the image processing module, is used to build an initial deep convolutional neural network model, input the lesion image into the initial deep convolutional neural network model for training, and obtain a target deep convolutional neural network model;

[0063] The disease resistance grade evaluation module is connected to the model building module and is used to evaluate and classify the images to be identified on the surfaces of litchi leaves or fruits at different stages and of different types based on the target deep convolutional neural network model, and obtain the disease resistance grade classification results.

[0064] Furthermore, the image acquisition module is an image acquisition device, specifically a high-resolution digital camera or an intelligent device with an integrated image sensor, such as a smart phone, tablet computer, etc., is used to shoot under natural light or controlled lighting conditions to ensure that the image is clear and shadow-free, so as to obtain a high-definition image of the litchi leaf or fruit. The image acquisition should cover the entire surface of the leaf or fruit to ensure that the characteristics of the lesions are fully recorded.

[0065] Furthermore, the image processing module includes a preprocessing unit and a feature extraction unit;

[0066] The preprocessing unit is used to perform image denoising, color correction, brightness and contrast adjustment on the collected images, eliminate the influence of light and shadow on the image quality, ensure that the features of the lesions are clearly visible, and obtain the preprocessed images;

[0067] The feature extraction unit is used to extract features from the preprocessed image to obtain a lesion feature image.

[0068] Further, the pre-processing unit includes a format conversion unit, a cropping and scaling unit, an image enhancement unit, and an image denoising unit;

[0069] The conversion format unit is used to convert the acquired images into a unified format, such as JPEG or PNG;

[0070] The cropping and scaling unit is used to crop the image by edge detection or threshold segmentation to remove irrelevant background, and to separate the leaf or fruit area from the background by computer vision technology, such as edge detection, threshold segmentation and other methods, to obtain the diseased area, and to scale the image of the diseased area to a uniform size, which is specifically 224×224 pixels in this embodiment;

[0071] The image enhancement unit is used to apply data enhancement techniques for rotation, flipping, color correction, brightness and contrast adjustment to increase sample diversity;

[0072] The image denoising unit is used to remove image noise through Gaussian filtering or median filtering.

[0073] Furthermore, the feature extraction unit develops specific image processing algorithms, such as gray-level co-occurrence matrix, morphological filtering, etc., based on color, texture and shape features to extract lesion features in order to accurately identify and extract lesion areas.

[0074] The feature extraction unit includes a color feature extraction unit, a texture feature extraction unit, a shape feature extraction unit, and a deep learning feature extraction unit;

[0075] The feature extraction unit is used to extract feature values ​​in RGB and HSV color spaces;

[0076] The texture feature extraction unit is used to extract texture features using methods such as gray level co-occurrence matrix (GLCM) or local binary pattern (LBP);

[0077] The shape feature extraction unit is used to analyze the shape, size and edge features of the lesions;

[0078] The deep learning feature extraction unit is used to automatically extract high-dimensional features through convolutional neural networks.

[0079] This embodiment develops an efficient feature extraction algorithm for accurate identification and quantification of anthrax lesions based on their specific morphology and texture, thereby achieving efficient evaluation.

[0080] Furthermore, an initial deep convolutional neural network model is constructed and trained to identify different stages and types of lesions. The model training dataset should contain a large number of labeled lesion images covering various disease development stages and lighting conditions.

[0081] Furthermore, the target deep convolutional neural network model is an improved ShuffleNet V2 model, which adopts hybrid atrous convolution and lightweight channel attention module to enhance feature extraction capability, uses Adam optimizer, sets the initial learning rate to 0.001, and adopts learning rate decay strategy, and uses cross entropy loss function as loss function.

[0082] Dataset Partitioning The dataset is divided into 70% training set, 15% validation set and 15% test set.

[0083] The input is the preprocessed image and its corresponding label (healthy / anthrax). The training cycle can be adjusted according to the loss and accuracy on the validation set, and usually tens to hundreds of cycles are required.

[0084] Evaluate the model performance on an independent test set through test set testing, and then perform error analysis to analyze the samples where the model misidentifies and find possible areas for improvement.

[0085] The present invention uses a deep convolutional neural network (CNN) model to identify and quantify the characteristics of anthracnose lesions on litchi leaves or fruits. The CNN model can automatically extract the characteristics of lesions in the image, such as shape, size, location and severity, thereby achieving accurate identification of lesions.

[0086] Further, the disease resistance grade evaluation module includes a lesion quantification analysis unit, a disease resistance grade standard establishment unit, a cluster analysis unit, and a grade evaluation unit;

[0087] The lesion quantitative analysis unit is used to identify and locate lesions in the image based on the target deep convolutional neural network model, obtain characteristic information of the lesions, and provide quantitative evaluation indicators for subsequent disease resistance evaluation based on the characteristic information;

[0088] The disease resistance level standard establishment unit is used to combine historical data and expert knowledge to set the threshold range of each disease resistance level, and divide the varieties into four disease resistance levels: high resistance, medium resistance, medium sensitivity, and high sensitivity. Each disease resistance level corresponds to a certain range of lesion area and lesion severity.

[0089] The cluster analysis unit is used to apply clustering algorithms, such as K-means and hierarchical clustering, to the lesion images, to analyze the quantitative data of lesions of different litchi varieties, and to classify the varieties into different disease resistance levels by using the intra-group linkage and squared Euclidean distance methods based on the lesion area and lesion severity.

[0090] This embodiment uses a cluster analysis method to process the image recognition results, and classifies the disease resistance of different litchi varieties according to the lesion quantification data. Through cluster analysis, the varieties can be objectively divided into different disease resistance levels such as high resistance, medium resistance, medium sensitivity, and high sensitivity, providing a scientific basis for the screening of disease-resistant varieties.

[0091] The grade evaluation unit is used to establish a decision tree model or a support vector machine model using a machine learning algorithm, input quantitative evaluation indicators, and output the corresponding disease resistance grade, thereby improving the automation and accuracy of disease resistance grade evaluation.

[0092] Based on the cluster analysis results of lesion quantification data, this embodiment proposes scientific and reasonable disease resistance grade classification standards and methods. The entire evaluation process, from image acquisition to lesion identification and then to disease resistance grade classification, is highly automated and intelligent, which greatly reduces human intervention, improves the efficiency and accuracy of the evaluation, and reduces subjective errors.

[0093] Furthermore, the characteristic information of the lesions includes, but is not limited to, the area, number, distribution, shape, location, and size of the lesions;

[0094] Quantitative evaluation indicators include but are not limited to accuracy, recall, and F1 score.

[0095] Furthermore, the system of the present invention further comprises:

[0096] Software interface design: Develop a user-friendly graphical interface to facilitate users to upload images, view analysis results and disease resistance grade reports.

[0097] Database and data management: Establish a database to store images and analysis results to facilitate data query, management and long-term tracking research.

[0098] Real-time monitoring and early warning system: Integrates Internet of Things technology to achieve remote image acquisition and instant analysis in field environments, providing real-time data support for disease early warning and prevention.

[0099] Cross-platform compatibility: Ensure that the system can run on a variety of devices, including PCs, mobile devices, etc., to meet the needs of users in different scenarios.

[0100] The technical solution of the present invention uses a variety of advanced algorithms such as image processing, deep learning and cluster analysis to build an efficient, accurate and convenient system for identifying the resistance level of litchi anthracnose, providing a scientific basis for the screening of litchi disease-resistant varieties and disease prevention and control. Through this system, the workload of manual evaluation can be greatly reduced, the accuracy and efficiency of identification can be improved, and strong technical support can be provided for the healthy development of the litchi industry.

[0101] In order to thoroughly demonstrate the effectiveness and superiority of the present invention "a method for rapid identification of litchi anthracnose resistance level based on image recognition", this section will describe in detail the experimental materials, methods, data processing flow, result analysis and control experiment design of the embodiment, aiming to verify the technical effect of the present invention through scientific and rigorous experiments.

[0102] Experimental materials: 31 different varieties of litchi in Maoming were selected as experimental objects, including but not limited to Bingli, Baila, Guiwei, Lingfengnuo, Heiye, Feizixiao, Baitangpeng, Fenghuangli, etc. The experiment was carried out during the litchi growing season to ensure the representativeness of the samples.

[0103] Experimental methods:

[0104] 1. Sample preparation: Select healthy and mature litchi leaves and fruits as test samples, and ensure that the samples are not affected by other diseases.

[0105] 2. Anthrax inoculation: Figure 2 and Figure 5 As shown, artificial inoculation was used to evenly apply the anthracnose pathogen solution to the leaves and fruit surfaces to simulate the natural infection process, ensuring that there were at least 10 replicate samples for each variety. The severity of the disease in the litchi leaf inoculation area is shown in Table 1, and the severity of the disease in the litchi fruit inoculation area is shown in Table 2.

[0106] Table 1

[0107]

[0108] Table 2

[0109]

[0110] 3. Image acquisition: On the 3rd, 7th, and 14th days after inoculation, use a high-resolution digital camera to take photos of the samples to ensure image clarity and complete lesion characteristics.

[0111] 4. Image preprocessing: Use image processing software to preprocess the collected images, including image denoising, color correction, brightness and contrast adjustment to ensure that the lesions are clearly visible.

[0112] 5. Image recognition and lesion analysis:

[0113] A deep convolutional neural network (CNN) model is used to identify and locate lesions in images.

[0114] Extract the size, shape, location and other characteristics of the lesions and quantify the lesion area.

[0115] 6. Cluster analysis: Based on the lesion area and severity, cluster analysis was performed using intra-group linkage and squared Euclidean distance methods to divide the varieties into four disease resistance levels: high resistance, medium resistance, medium sensitivity, and high sensitivity. Figure 3 and Figure 6 As shown, A on the left is the cluster analysis based on the lesion area, and B on the right is the cluster analysis based on the disease index; among them, I: immune; HR: high resistance; MR: moderate resistance; MS: moderate infection; HS: high infection.

[0116] Data processing:

[0117] 1. Statistical analysis: Use SPSS or R language statistical software to perform descriptive statistical analysis on the lesion area data and calculate statistical quantities such as mean value and standard deviation.

[0118] 2. Cluster analysis: Cluster analysis was performed using the squared Euclidean distance of 3.0 as the dividing point to determine the criteria for different disease resistance levels.

[0119] Results and Analysis:

[0120] 1. Disease resistance level classification: Figure 4 and Figure 7 As shown in the figure, based on the clustering results, the classification of varieties with different disease resistance levels was clarified. For example, Bingli and Baila were classified as highly resistant varieties, Guiwei and Lingfengnuo were medium-resistant varieties, Feizishao and Heiye were medium-susceptible varieties, and Baitangpeng and Fenghuangli were highly susceptible varieties.

[0121] 2. Disease resistance comparison: The comparison and analysis of the development speed and area change trend of the lesions of varieties with different disease resistance levels confirmed the effectiveness and rationality of the method of the present invention. The resistance identification standards of inoculated litchi leaves are shown in Table 3, and the resistance identification standards of inoculated litchi fruits are shown in Table 4.

[0122] Table 3

[0123]

[0124] Table 4

[0125]

[0126]

[0127] Control experiment:

[0128] A control group was set up, and the disease resistance level of the same group of experimental samples was evaluated using the traditional manual visual evaluation method. The evaluation results of the method of the present invention and the traditional method were compared and analyzed, and evaluated from the aspects of evaluation efficiency, accuracy and consistency, proving that the method of the present invention can improve the evaluation efficiency while maintaining or even surpassing the evaluation accuracy of the traditional method.

[0129] Result analysis:

[0130] Through statistical analysis of experimental data and comparison of clustering results, it is confirmed that the image recognition method of the present invention can quickly and accurately identify the anthracnose resistance level of different litchi varieties. Compared with traditional manual visual evaluation, the method of the present invention has higher efficiency and consistency, effectively solves the subjective problem existing in manual evaluation, and provides a scientific basis for the screening of disease-resistant litchi varieties and disease prevention and control.

[0131] in conclusion:

[0132] The invention proposes a method for rapid identification of the disease resistance level of litchi anthracnose based on image recognition. The method automatically identifies anthracnose lesions on litchi leaves and fruits through intelligent image processing technology. Images of litchi leaves and fruits are collected and analyzed using machine learning or deep learning models, and finally the identification results are output. After detailed experimental verification, it is proved that the method has significant advantages in improving evaluation efficiency and ensuring evaluation accuracy, and provides strong technical support for disease management and variety improvement in the litchi industry.

[0133] Embodiment 2

[0134] This embodiment also provides a method for quickly identifying the disease resistance level of litchi anthracnose based on image recognition, comprising the following steps:

[0135] Acquiring an image covering the entire surface of a litchi leaf or fruit, preprocessing the image, and then performing feature extraction to obtain a disease spot image;

[0136] Constructing an initial deep convolutional neural network model, inputting the lesion image into the initial deep convolutional neural network model for training, and obtaining a target deep convolutional neural network model;

[0137] Based on the target deep convolutional neural network model, the images to be identified on the surfaces of litchi leaves or fruits at different stages and of different types are evaluated and classified to obtain disease resistance grade classification results.

[0138] The above are only preferred specific implementations of the present application, but the protection scope of the present application is not limited thereto. Any changes or substitutions that can be easily thought of by a person skilled in the art within the technical scope disclosed in the present application should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

Claims

1. A rapid identification system for the disease resistance level of litchi anthracnose based on image recognition, characterized in that: include: An image acquisition module is used to acquire images covering the entire surface of litchi leaves or fruits; An image processing module, connected to the image acquisition module, is used to perform feature extraction on the image after preprocessing to obtain a lesion image; A model building module, connected to the image processing module, for building an initial deep convolutional neural network model, inputting the lesion image into the initial deep convolutional neural network model for training, and obtaining a target deep convolutional neural network model; The disease resistance grade evaluation module is connected to the model building module and is used to evaluate and classify the to-be-identified images of litchi leaves or fruit surfaces at different stages and of different types based on the target deep convolutional neural network model, and obtain disease resistance grade classification results.

2. The rapid identification system for the disease resistance grade of litchi anthracnose based on image recognition according to claim 1 is characterized in that: The image processing module includes a preprocessing unit and a feature extraction unit; The preprocessing unit is used to perform image denoising, color correction, brightness and contrast adjustment on the collected image, eliminate the influence of light and shadow on the image quality, and obtain a preprocessed image; The feature extraction unit is used to extract features from the preprocessed image to obtain a lesion feature image.

3. The rapid identification system for the disease resistance grade of litchi anthracnose based on image recognition according to claim 2 is characterized in that: The preprocessing unit includes a format conversion unit, a cropping and scaling unit, an image enhancement unit, and an image denoising unit; The format conversion unit is used to convert the collected images into a unified format; The cropping and scaling unit is used to crop the image by edge detection or threshold segmentation, remove irrelevant background, separate the leaf or fruit area from the background, obtain the diseased spot area, and scale the image of the diseased spot area to a uniform size; The image enhancement unit is used to apply data enhancement technology to perform rotation, flipping, color correction, brightness and contrast adjustment; The image denoising unit is used to remove image noise through Gaussian filtering or median filtering.

4. The rapid identification system for the disease resistance grade of litchi anthracnose based on image recognition according to claim 2 is characterized in that: The feature extraction unit includes a color feature extraction unit, a texture feature extraction unit, a shape feature extraction unit, and a deep learning feature extraction unit; The feature extraction unit is used to extract feature values ​​in RGB and HSV color spaces; The texture feature extraction unit is used to extract texture features using a gray level co-occurrence matrix or a local binary pattern method; The shape feature extraction unit is used to analyze the shape feature, size feature and edge feature of the lesion; The deep learning feature extraction unit is used to automatically extract high-dimensional features through a convolutional neural network.

5. The rapid identification system for litchi anthracnose resistance grade based on image recognition according to claim 1 is characterized in that: The target deep convolutional neural network model is an improved ShuffleNetV2 model, which adopts hybrid hole convolution and lightweight channel attention module to enhance feature extraction capability, uses Adam optimizer, sets the initial learning rate to 0.001, and adopts learning rate decay strategy, and adopts cross entropy loss function as loss function.

6. The rapid identification system for litchi anthracnose resistance grade based on image recognition according to claim 1, characterized in that: The disease resistance grade evaluation module includes a lesion quantification analysis unit, a disease resistance grade standard establishment unit, a cluster analysis unit, and a grade evaluation unit; The lesion quantitative analysis unit is used to identify and locate lesions in the image based on the target deep convolutional neural network model, obtain characteristic information of the lesions, and provide a quantitative evaluation index for disease resistance evaluation based on the characteristic information; The disease resistance level standard establishment unit is used to combine historical data and expert knowledge to set the threshold range of each disease resistance level, and divide the varieties into four disease resistance levels: high resistance, medium resistance, medium sensitivity, and high sensitivity. Each disease resistance level corresponds to a certain range of lesion area and lesion severity. The cluster analysis unit is used to analyze the quantitative data of lesions of different litchi varieties using a clustering algorithm on the lesion images, and to classify the varieties into different disease resistance levels by using the intra-group linkage and squared Euclidean distance method according to the lesion area and lesion severity; The grade evaluation unit is used to establish a decision tree model or a support vector machine model using a machine learning algorithm, input a quantitative evaluation index, and output a corresponding disease resistance grade.

7. The rapid identification system for litchi anthracnose resistance grade based on image recognition according to claim 6 is characterized in that: The characteristic information of the lesions includes but is not limited to the area, number, distribution, shape, position, and size of the lesions.

8. The rapid identification system for litchi anthracnose resistance grade based on image recognition according to claim 6 is characterized in that: The quantitative evaluation indicators include but are not limited to accuracy, recall rate, and F1 score.

9. A method for rapid identification of litchi anthracnose resistance grade based on image recognition, characterized in that: The following steps are involved: Acquiring an image covering the entire surface of a litchi leaf or fruit, preprocessing the image, and then performing feature extraction to obtain a disease spot image; Constructing an initial deep convolutional neural network model, inputting the lesion image into the initial deep convolutional neural network model for training, and obtaining a target deep convolutional neural network model; Based on the target deep convolutional neural network model, the images to be identified on the surfaces of litchi leaves or fruits at different stages and of different types are evaluated and classified to obtain disease resistance grade classification results.

Images