Fruit plant leaf lesion detection method and system based on deep learning
Through the deep learning-based detection method for leaf lesions of fruit plants, the problem of insufficient detection efficiency and accuracy in the existing technology is solved, and the detection accuracy and efficiency are improved. It is suitable for large-scale agricultural production environments.
Patent Information
- Application Number
- CN202411892585.2
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-12-20
- Publication Date
- 2025-04-15
AI Technical Summary
The existing methods for leaf disease detection of fruit plants rely on manual observation and empirical judgment, resulting in insufficient detection efficiency and accuracy, and it is difficult to achieve high-precision identification under complex and diverse lesions, and cannot be applied to large-scale agricultural production environments.
Using deep learning-based detection methods, we automatically identify and classify leaf lesions of fruit plants by acquiring high-resolution images, performing image preprocessing, building and training deep learning models.
It improves detection efficiency and accuracy, reduces labor costs, enhances the detection accuracy of different lesion types, is suitable for large-scale agricultural production environments, reduces losses caused by disease spread, and improves crop yield and quality.
Smart Images
Figure CN119832421A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of fruit plant leaf lesion detection, and in particular to a fruit plant leaf lesion detection method and system based on deep learning. Background Art
[0002] Fruit plant leaf lesions may be caused by a variety of reasons, including fungi, bacteria, viruses, pests, etc. In modern agricultural production, early detection and management of plant leaf diseases are essential to maintain crop health and yield.
[0003] At present, most of the existing disease detection methods rely on manual observation and empirical judgment, resulting in serious lack of detection efficiency and accuracy. It is not only time-consuming and labor-intensive, but also makes the detection results easily affected by human factors. In addition, it is generally impossible to achieve high-precision identification when faced with complex and diverse lesions. At the same time, it is also not suitable for large-scale agricultural production environments. Therefore, it is necessary to design a fruit plant leaf lesion detection method and system based on deep learning. Summary of the invention
[0004] The purpose of the present invention is to overcome the shortcomings of the prior art, in order to better and effectively solve the existing disease detection methods, which mostly rely on manual observation and experience judgment, resulting in serious lack of detection efficiency and accuracy, not only time-consuming and labor-intensive, but also make the detection results easily affected by human factors, and generally cannot achieve high-precision recognition when facing complex and diverse lesion conditions, and are also not applicable to problems in large-scale agricultural production environments. A fruit plant leaf lesion detection method and system based on deep learning are provided, which realizes that the fruit plant leaf lesion detection method and system have the function of automatically identifying and classifying lesions on the surface of fruit plant leaves using a deep learning model, and also realizes that the normal and abnormal features of the fruit plant leaf surface can be automatically learned and labeled, and by training a large-scale fruit plant leaf image data set, the deep learning model can gradually improve the detection accuracy of different lesion types, which not only improves the detection efficiency and accuracy, but also reduces labor costs.
[0005] In order to achieve the above object, the technical solution adopted by the present invention is:
[0006] A method for detecting leaf lesions of fruit plants based on deep learning comprises the following steps:
[0007] Step A, collecting high-resolution images of leaves of fruit plants to obtain collected images;
[0008] Step B, performing image preprocessing on the collected image to obtain a preprocessed image;
[0009] Step C, constructing a deep learning model, and then training and verifying the deep learning model to obtain a verified deep learning model;
[0010] Step D, using the verified deep learning model to extract features and classify lesion areas from the preprocessed image to obtain detection results;
[0011] Step E, recording and feeding back the detection results, completing the fruit plant leaf lesion detection operation.
[0012] The aforementioned method for detecting leaf lesions of fruit plants based on deep learning, step A, collects high-resolution images of leaves of fruit plants to obtain collected images, specifically using high-resolution camera equipment to capture clear leaf images under different lighting conditions, and the collected images specifically cover leaves of different types of fruit plants in various health conditions and pathological conditions.
[0013] In the above-mentioned fruit plant leaf lesion detection method based on deep learning, step B performs image preprocessing on the collected image to obtain a preprocessed image, wherein the image preprocessing includes denoising, color correction, random cropping, rotation, horizontal flipping, brightness adjustment and image normalization, wherein the image normalization specifically adjusts the collected image I to a fixed size W×H to obtain a preprocessed image I′, as shown in formula (1),
[0014] I′=Resize(I,W,H) (1)
[0015] Resize is to scale and resize the image I so that the processed image I′ has a set width W and height H.
[0016] The above-mentioned method for detecting leaf lesions of fruit plants based on deep learning, step C, constructing a deep learning model, and then training and verifying the deep learning model to obtain a verified deep learning model, the specific steps are as follows:
[0017] Step C1, constructing a deep learning model, wherein the deep learning model comprises a backbone network and a Neck network, wherein the backbone network is used to extract multi-scale features in an image and generate a multi-scale feature map, and the Neck network is used to perform feature fusion on the multi-scale feature map and generate feature maps at different levels. The specific steps are as follows:
[0018] Step C11, construct the backbone network, as shown in formula (2),
[0019] F = Conv2D(I′,K,S,P)
[0020] Among them, F is a nonlinear activation function, Conv2D is a two-dimensional convolution operation, K is the convolution kernel size, S is the stride, and P is padding;
[0021] Step C12, construct the Neck network, as shown in formula (3),
[0022] F fuse =Conact(F low ,F high ) (3)
[0023] Among them, F fuse is the fused feature, F low is the high-resolution low-level feature, F high It is a low-resolution high-level feature;
[0024] Step C2: training the constructed deep learning model. The specific training process is shown in formula (4):
[0025] X1=BN(Conv 1*1 (X)),
[0026] X K =BN(Conv K *K(X)),
[0027] Output=F(X1+BN(Conv K * K (X1)+BN(AVG(X1)+X K ) (4)
[0028] Among them, X is the input feature, X1 is the result after the convolution operation with a convolution kernel of 1*1, and X K is the result after the convolution operation with a convolution kernel of K*K, BN is the normalization operation, Conv 1*1 Convolution operation with a convolution kernel of 1*1, Conv K*K It is a convolution operation with a convolution kernel of K*K, Output is the output feature, and AVG represents average pooling;
[0029] Step C3, verify the trained deep learning model, including bounding box verification, target confidence verification and category probability verification. The specific steps are as follows:
[0030] Step C31, perform bounding box verification on the trained deep learning model, specifically using the anchor-free method to calculate the center point and width and height offset of the bounding box, as shown in formula (5):
[0031]
[0032] Among them, x c is the horizontal coordinate of the center point of the final bounding box, y cis the ordinate of the center point of the final bounding box, σ is the Sigmoid function, and t x and t y are the offsets of the predicted point relative to the reference point in the x and y directions, c x is the horizontal coordinate of the reference point, c y is the ordinate of the reference point, ω is the final bounding box width, h is the final bounding box height, p ω is the reference frame width, p h is the reference frame height, and All are exponential mappings;
[0033] Step C32, verifying the target confidence of the trained deep learning model, wherein the target confidence indicates whether the target box contains an object, and the target confidence verification process specifically converts the confidence measurement result t c The value is converted into a confidence score C using the Sigmoid method, as shown in formula (6):
[0034] C=σ(t c )(6);
[0035] Step C33, verify the category probability of the trained deep learning model, specifically, calculate the category probability P for each category, as shown in formula (7),
[0036] P = Softmax(s) (7)
[0037] Among them, s is the score vector of each category, and Softmax is converted into probability distribution.
[0038] In the aforementioned method for detecting leaf lesions of fruit plants based on deep learning, step D uses a verified deep learning model to perform feature extraction and lesion area classification on the preprocessed image to obtain a detection result, wherein the detection result includes detection time, lesion type, lesion area size, lesion severity and recommended treatment measures.
[0039] In the aforementioned method for detecting leaf lesions of fruit plants based on deep learning, step E records and feeds back the detection results to complete the detection operation of leaf lesions of fruit plants. Specifically, the detection results are recorded and saved in a database to form a historical record of plant lesions.
[0040] A fruit plant leaf lesion detection system based on deep learning comprises an image acquisition module, an image preprocessing module, a model building module, an image detection module and a data recording module, wherein the image acquisition module is used to acquire high-resolution images of fruit plant leaves and obtain acquired images; the image preprocessing module is used to perform image preprocessing on the acquired images and obtain preprocessed images; the model building module is used to construct a deep learning model and train and verify the deep learning model to obtain a verified deep learning model; the image detection module is used to use the verified deep learning model to perform feature extraction and lesion area classification on the preprocessed images and obtain detection results; the data recording module is used to record and feedback the detection results to complete the fruit plant leaf lesion detection operation.
[0041] The beneficial effects of the present invention are as follows: a method and system for detecting leaf lesions of fruit plants based on deep learning of the present invention first collects high-resolution images of fruit plant leaves and obtains collected images, then performs image preprocessing on the collected images and obtains preprocessed images, then builds a deep learning model, then trains and verifies the deep learning model and obtains a verified deep learning model, then uses the verified deep learning model to extract features and classify lesion areas of the preprocessed images and obtain detection results, then records and feeds back the detection results to complete the detection of leaf lesions of fruit plants; effectively realizing that the method and system for detecting leaf lesions of fruit plants have the advantages of using deep learning The deep learning model has the function of automatically identifying and classifying lesions on the surface of fruit plant leaves. It also improves the generalization ability of the deep learning model by using image acquisition equipment to obtain image data of fruit plant leaves and preprocessing the image data. It can also automatically learn the normal and abnormal features of the fruit plant leaf surface and label them. By training a large-scale fruit plant leaf image dataset, the deep learning model can gradually improve the detection accuracy of different lesion types, which not only improves the detection efficiency and accuracy, but also reduces labor costs. At the same time, it also reduces the losses caused by the spread of diseases, increases crop yield and quality, and realizes the intelligence, automation and efficiency of agricultural production. BRIEF DESCRIPTION OF THE DRAWINGS
[0042] Figure 1 It is an overall flow chart of a method for detecting leaf lesions of fruit plants based on deep learning of the present invention;
[0043] Figure 2 It is a schematic diagram of the detection principle of a fruit plant leaf lesion detection system based on deep learning of the present invention;
[0044] Figure 3 It is a schematic diagram of the structure of a deep learning model of a fruit plant leaf lesion detection method based on deep learning of the present invention;
[0045] Figure 4 It is a detection function schematic diagram of a fruit plant leaf lesion detection method based on deep learning of the present invention;
[0046] Figure 5 It is a schematic diagram of the detection results of a fruit plant leaf lesion detection method based on deep learning of the present invention. DETAILED DESCRIPTION
[0047] The present invention will be further described below in conjunction with the accompanying drawings.
[0048] like Figure 1 As shown, a fruit plant leaf lesion detection method based on deep learning of the present invention comprises the following steps:
[0049] Step A, collecting high-resolution images of fruit plant leaves to obtain collected images, specifically using a high-resolution camera to capture clear leaf images under different lighting conditions, and the collected images specifically cover leaves of different types of fruit plants in various health conditions and pathological conditions.
[0050] Step B, performing image preprocessing on the collected image to obtain a preprocessed image, wherein the image preprocessing includes denoising, color correction, random cropping, rotation, horizontal flipping, brightness adjustment and image normalization, wherein image normalization specifically adjusts the collected image I to a fixed size W×H to obtain a preprocessed image I′, as shown in formula (1),
[0051] I′=Resize(I,W,H) (1)
[0052] Resize is to scale and resize the image I so that the processed image I′ has a set width W and height H.
[0053] like Figure 3 As shown, step C is to build a deep learning model, then train and verify the deep learning model to obtain a verified deep learning model. The specific steps are as follows:
[0054] Step C1, constructing a deep learning model, wherein the deep learning model comprises a backbone network and a Neck network, wherein the backbone network is used to extract multi-scale features in an image and generate a multi-scale feature map, and the Neck network is used to perform feature fusion on the multi-scale feature map and generate feature maps at different levels. The specific steps are as follows:
[0055] Step C11, construct the backbone network, as shown in formula (2),
[0056] F = Conv2D(I′,K,S,P)
[0057] Among them, F is a nonlinear activation function, Conv2D is a two-dimensional convolution operation, K is the convolution kernel size, S is the stride, and P is padding;
[0058] Step C12, construct the Neck network, as shown in formula (3),
[0059] F fuse =Conact(F low ,F high ) (3)
[0060] Among them, F fuse is the fused feature, F low is the high-resolution low-level feature, F high It is a low-resolution high-level feature;
[0061] Step C2: training the constructed deep learning model. The specific training process is shown in formula (4):
[0062] X1=BN(Conv 1*1 (X)),
[0063] X K =BN(Conv K*K (X)),
[0064] Output=F(X1+BN(Conv K*K (X1)+BN(AVG(X1)+X K ) (4)
[0065] Among them, X is the input feature, X1 is the result after the convolution operation with a convolution kernel of 1*1, and X K is the result after the convolution operation with a convolution kernel of K*K, BN is the normalization operation, Conv 1*1 Convolution operation with a convolution kernel of 1*1, Conv K*K It is a convolution operation with a convolution kernel of K*K, Output is the output feature, and AVG represents average pooling;
[0066] Step C3, verify the trained deep learning model, including bounding box verification, target confidence verification and category probability verification. The specific steps are as follows:
[0067] Step C31, perform bounding box verification on the trained deep learning model, specifically using the anchor-free method to calculate the center point and width and height offset of the bounding box, as shown in formula (5):
[0068]
[0069] Among them, x c is the horizontal coordinate of the center point of the final bounding box, yc is the ordinate of the center point of the final bounding box, σ is the Sigmoid function, and t x and t y are the offsets of the predicted point relative to the reference point in the x and y directions, c x is the horizontal coordinate of the reference point, c y is the ordinate of the reference point, ω is the final bounding box width, h is the final bounding box height, p ω is the reference frame width, p h is the reference frame height, and All are exponential mappings;
[0070] Step C32, verifying the target confidence of the trained deep learning model, wherein the target confidence indicates whether the target box contains an object, and the target confidence verification process specifically converts the confidence measurement result t c The value is converted into a confidence score C using the Sigmoid method, as shown in formula (6):
[0071] C=σ(t c )(6);
[0072] Step C33, verify the category probability of the trained deep learning model, specifically, calculate the category probability P for each category, as shown in formula (7),
[0073] P = Softmax(s) (7)
[0074] Among them, s is the score vector of each category, and Softmax is converted into probability distribution.
[0075] like Figure 4 As shown, in step D, the verified deep learning model is used to extract features and classify the lesion area of the preprocessed image to obtain a detection result, which includes the detection time, lesion type, lesion area size, lesion severity and recommended treatment measures.
[0076] like Figure 5 As shown, step E records and feeds back the test results to complete the fruit plant leaf lesion detection operation, specifically recording and saving the test results to a database to form a plant lesion history record.
[0077] like Figure 2As shown, a fruit plant leaf lesion detection system based on deep learning includes an image acquisition module, an image preprocessing module, a model building module, an image detection module and a data recording module, wherein the image acquisition module is used to acquire high-resolution images of fruit plant leaves and obtain acquired images; the image preprocessing module is used to perform image preprocessing on the acquired images and obtain preprocessed images; the model building module is used to construct a deep learning model and train and verify the deep learning model to obtain a verified deep learning model; the image detection module is used to use the verified deep learning model to perform feature extraction and lesion area classification on the preprocessed image and obtain detection results; the data recording module is used to record and feedback the detection results to complete the fruit plant leaf lesion detection operation.
[0078] In summary, the present invention provides a method and system for detecting leaf lesions of fruit plants based on deep learning. First, high-resolution images of fruit plant leaves are collected to obtain collected images. Then, image preprocessing is performed on the collected images to obtain preprocessed images. Then, a deep learning model is constructed. Then, the deep learning model is trained and verified to obtain a verified deep learning model. Subsequently, the verified deep learning model is used to extract features and classify lesion areas of the preprocessed images to obtain detection results. Then, the detection results are recorded and fed back to complete the detection of leaf lesions of fruit plants. The method and system for detecting leaf lesions of fruit plants have the advantages of using a deep learning model. It has the function of automatically identifying and classifying lesions on the surface of fruit plant leaves, and improves the generalization ability of the deep learning model by using image acquisition equipment to obtain image data of fruit plant leaves and preprocessing the image data. It can also automatically learn the normal and abnormal features of the surface of fruit plant leaves and label them. By training a large-scale fruit plant leaf image data set, the deep learning model can gradually improve the detection accuracy of different types of lesions, which not only improves the detection efficiency and accuracy, but also reduces labor costs. At the same time, it also reduces the losses caused by the spread of diseases, increases crop yield and quality, and realizes the intelligence, automation and efficiency of agricultural production.
[0079] The above shows and describes the basic principles, main features and advantages of the present invention. It should be understood by those skilled in the art that the present invention is not limited to the above embodiments. The above embodiments and descriptions are only for explaining the principles of the present invention. Without departing from the spirit and scope of the present invention, the present invention may have various changes and improvements, which fall within the scope of the present invention to be protected. The scope of protection of the present invention is defined by the attached claims and their equivalents.
Claims
1. A method for detecting leaf lesions of fruit plants based on deep learning, characterized in that: The following steps are included: Step A, collecting high-resolution images of leaves of fruit plants to obtain collected images; Step B, performing image preprocessing on the collected image to obtain a preprocessed image; Step C, constructing a deep learning model, and then training and verifying the deep learning model to obtain a verified deep learning model; Step D, using the verified deep learning model to extract features and classify lesion areas from the preprocessed image to obtain detection results; Step E, recording and feeding back the detection results, completing the fruit plant leaf lesion detection operation.
2. The method for detecting leaf lesions of fruit plants based on deep learning according to claim 1, characterized in that: Step A, collecting high-resolution images of fruit plant leaves to obtain collected images, specifically using a high-resolution camera to capture clear leaf images under different lighting conditions, and the collected images specifically cover leaves of different types of fruit plants in various health conditions and pathological conditions.
3. The method for detecting leaf lesions of fruit plants based on deep learning according to claim 2, characterized in that: Step B, performing image preprocessing on the collected image to obtain a preprocessed image, wherein the image preprocessing includes denoising, color correction, random cropping, rotation, horizontal flipping, brightness adjustment and image normalization, wherein image normalization specifically adjusts the collected image I to a fixed size W×H to obtain a preprocessed image I′, as shown in formula (1), I′=Resize(I,W,H) (1) Resize is to scale and resize the image I so that the processed image I′ has a set width W and height H.
4. The method for detecting leaf lesions of fruit plants based on deep learning according to claim 3, characterized in that: Step C: construct a deep learning model, then train and verify the deep learning model to obtain a verified deep learning model. The specific steps are as follows: Step C1, constructing a deep learning model, wherein the deep learning model comprises a backbone network and a Neck network, wherein the backbone network is used to extract multi-scale features in an image and generate a multi-scale feature map, and the Neck network is used to perform feature fusion on the multi-scale feature map and generate feature maps at different levels. The specific steps are as follows: Step C11, construct the backbone network, as shown in formula (2), F = Conv2D(I′,K,S,P) Among them, F is a nonlinear activation function, Conv2D is a two-dimensional convolution operation, K is the convolution kernel size, S is the stride, and P is padding; Step C12, construct the Neck network, as shown in formula (3), F fuse =Conact(F low ,F high )(3) Among them, F fuse is the fused feature, F low is the high-resolution low-level feature, F high It is a low-resolution high-level feature; Step C2: training the constructed deep learning model. The specific training process is shown in formula (4): X1=BN(Conv 1*1 (X)), X K =BN(Conv K*K (X)), Output=F(X1+BN(Conv K*K (X1)+BN(AVG(X1)+X K ) (4) Among them, X is the input feature, X1 is the result after the convolution operation with a convolution kernel of 1*1, and X K is the result after the convolution operation with a convolution kernel of K*K, BN is the normalization operation, Conv 1*1 Convolution operation with a convolution kernel of 1*1, Conv K*K It is a convolution operation with a convolution kernel of K*K, Output is the output feature, and AVG represents average pooling; Step C3, verify the trained deep learning model, including bounding box verification, target confidence verification and category probability verification. The specific steps are as follows: Step C31, perform bounding box verification on the trained deep learning model, specifically using the anchor-free method to calculate the center point and width and height offset of the bounding box, as shown in formula (5): Among them, x c is the horizontal coordinate of the center point of the final bounding box, y c is the ordinate of the center point of the final bounding box, σ is the Sigmoid function, and t x and t y are the offsets of the predicted point relative to the reference point in the x and y directions, c x is the horizontal coordinate of the reference point, c y is the ordinate of the reference point, ω is the final bounding box width, h is the final bounding box height, p ω is the reference frame width, p h is the reference frame height, and All are exponential mappings; Step C32, verifying the target confidence of the trained deep learning model, wherein the target confidence indicates whether the target box contains an object, and the target confidence verification process specifically converts the confidence measurement result t c The value is converted into a confidence score C using the Sigmoid method, as shown in formula (6): C=σ(t c )(6); Step C33, verify the category probability of the trained deep learning model, specifically, calculate the category probability P for each category, as shown in formula (7), P = Softmax(s) (7) Among them, s is the score vector of each category, and Softmax is converted into probability distribution.
5. The method for detecting leaf lesions of fruit plants based on deep learning according to claim 4, characterized in that: Step D: Use the verified deep learning model to extract features and classify the lesion area of the preprocessed image to obtain the detection results, which include the detection time, lesion type, lesion area size, lesion severity and recommended treatment measures.
6. The method for detecting leaf lesions of fruit plants based on deep learning according to claim 5, characterized in that: Step E, recording and feeding back the test results, completing the fruit plant leaf lesion detection operation, specifically recording and saving the test results to a database to form a plant lesion history record.
7. A fruit plant leaf lesion detection system based on deep learning, wherein the specific detection process of the fruit plant leaf lesion detection system is based on the fruit plant leaf lesion detection method obtained according to any one of claims 1 to 6, characterized in that: It includes an image acquisition module, an image preprocessing module, a model building module, an image detection module and a data recording module, wherein the image acquisition module is used to acquire high-resolution images of fruit plant leaves and obtain acquired images; The image preprocessing module is used to perform image preprocessing on the collected image and obtain the preprocessed image; The model building module is used to build a deep learning model and train and verify the deep learning model to obtain a verified deep learning model; The image detection module is used to extract features and classify lesion areas of the preprocessed image using a verified deep learning model and obtain detection results; The data recording module is used to record and feed back the detection results so as to complete the fruit plant leaf lesion detection operation.