Diagnosis and control method for diseases and insect pests of transplanted red beans

Abstract

The invention discloses a method for diagnosing, preventing and treating diseases and insect pests of transplanted red beans. The method comprises the following steps: deploying video monitoring and collecting and planting leaf images of the red beans; carrying out adaptive wavelet threshold denoising on the collected red bean leaf image; performing two-channel fusion segmentation on the de-noised red bean leaf image to obtain a red bean pest and disease damage area image; constructing a disease and pest diagnosis model based on a CNN neural network; inputting the red bean pest and disease damage area image into a pest and disease damage diagnosis model and outputting a pest and disease damage probability; and when the pest probability exceeds a threshold, generating a control scheme. According to the invention, the video monitoring image is automatically collected, denoising, segmentation and pest and disease prediction processing are carried out, the focus area can be better positioned, the characteristics of different pests and diseases can be grasped, and the prevention and control scheme is dynamically generated according to different pests and diseases.

Description

Technical Field

[0001] The invention relates to the field of disease and insect pest diagnosis, and in particular to a method for diagnosing and preventing diseases and insect pests of transplanted rock-grown red beans. Background Art

[0002] The rock red bean tree has a beautiful shape, an evergreen crown, and bright seeds, making it an ornamental tree species for garden greening. With a straight trunk, high-quality wood, and beautiful texture, it is a timber species of the redwood family. The seeds have the effects of clearing heat and detoxifying, reducing swelling and removing dampness, and are an important medicinal material. The reasons for its endangered status include the number of mature plants, flowering and fruiting, seed characteristics, reproductive capacity and other inherent characteristics, which are often the primary factors restricting the population's own reproduction. Although some rare and endangered plants can produce fruit, due to the hard and dense seed coat, conventional sowing makes it difficult for them to germinate and seedling, resulting in a weak natural regeneration ability and a rare state. At the same time, the growth, reproduction and regeneration of rock red bean are also faced with multiple external factors, including environmental changes, interspecific competition, and human activities.

[0003] With the rapid development of information technology, AI, the Internet of Things, big data, and mobile internet, a growing number of technologies and products are being applied to plant growth monitoring and pest and disease identification and diagnosis. Refined production and pest and disease management based on the concept of precision agriculture are gradually becoming the mainstream of agricultural development. Currently, rapid pest and disease diagnosis is mainly achieved through contour recognition and deep learning. However, traditional contour recognition methods rely on manual feature extraction and are difficult to meet the needs of automation.

[0004] For example, the Chinese patent with the authorization number CN106708782B discloses a method for regional pest detection, diagnosis and discrimination based on wavelet analysis. The method includes the following steps: the first step is to extract and sample the experimental data of Populus euphratica pests; the second step is to use wavelet transform to perform preliminary analysis on the sampled data; the third step is to define and construct the NDVI time spectrum curve; the fourth step is to filter the noise of the NDVI time spectrum curve; the fifth step is to enhance and separate the Populus euphratica pests; the sixth step is to establish a signal detection model for Populus euphratica leaf-feeding pests; the seventh step is to perform real-time dynamic detection of the detected pest signals. This invention is based on the principles and physical basis of remote sensing monitoring of forest pests and diseases. With the help of NDVI time series data, it defines and constructs the Populus euphratica NDVI time spectrum curve containing pest information, obtains the evolution characteristics and laws of NDVI over time, effectively monitors plant growth and pests and diseases, and prevents the gradual weakening of Populus euphratica growth.

[0005] For example, Chinese patent application CN113688959B discloses an artificial intelligence-based plant disease and insect pest diagnosis method and system. The method includes the following steps: obtaining disease and style features of each labeled plant sample image based on a color and texture distribution map; training a variational autoencoder using the labeled plant sample images to obtain a Gaussian model corresponding to each plant sample image to be labeled; assigning labels to each plant sample image to be labeled based on the Gaussian model to obtain label data corresponding to each plant sample image to be labeled; training a target network based on the labeled plant sample images, the label data corresponding to each labeled plant sample image, the plant sample images to be labeled, and the label data corresponding to each plant sample image to be labeled; and inputting the target plant image into the trained target network to obtain a diagnostic result corresponding to the target plant image. This invention can improve the accuracy of the target network in diagnosing plant diseases and insect pests.

[0006] Due to the diverse types of leaf diseases and insect pests in transplanted rock bean, the image features collected are also different. Traditional lesion area contour recognition relies on manual extraction, and due to the wide variety and complex morphology of pests and diseases, traditional contour recognition methods cannot meet the requirements of accurate diagnosis. Summary of the Invention

[0007] The technical problem to be solved by the present invention is to provide a method for diagnosing and preventing diseases and insect pests of transplanted rock-grown red beans in response to the deficiencies of the existing technology.

[0008] In order to achieve the above object, the technical solution adopted by the present invention is:

[0009] The diagnosis and control methods of diseases and pests of transplanted rock red beans include the following steps:

[0010] Deploy video surveillance to collect images of rock red bean leaves;

[0011] Adaptive wavelet threshold denoising was performed on the collected images of Echinops lithophylla leaves;

[0012] The denoised red bean leaf image is subjected to dual-channel fusion segmentation to obtain the red bean disease and insect pest area image;

[0013] Construct a pest and disease diagnosis model based on CNN neural network;

[0014] The image of the pest and disease area of ​​rock red bean is input into the pest and disease diagnosis model and the pest and disease probability is output;

[0015] When the probability of pests and diseases exceeds the threshold, a prevention and control plan is generated.

[0016] Furthermore, the adaptive wavelet threshold denoising is performed on the collected transplanted rock red bean leaf image, specifically comprising the following steps:

[0017] Perform three-layer wavelet decomposition on the collected leaf images of red bean:

[0018] The three-layer wavelet decomposition is divided into subbands and adaptive thresholds are calculated according to subband categories, wherein the subband categories include: high-frequency noise dominant category, horizontal lesion edge category, vertical lesion edge category, and low-frequency signal retention category;

[0019] Improve the soft threshold function to selectively suppress the high-frequency coefficients after wavelet decomposition;

[0020] The wavelet was reconstructed to generate the denoised image of the leaf of Echinops lithophylla.

[0021] Furthermore, the three-layer wavelet decomposition of the collected leaf image of the red bean is performed, specifically including:

[0022] The first layer is decomposed into: LL1, HL1, LH1 and HH1, where LL1 represents the low-frequency approximation subband; HL1, LH1 and HH1 represent the high-frequency detail subbands in the horizontal, vertical and diagonal directions respectively;

[0023] The LL1 subband is decomposed into the following two layers: LL2, HL2, LH2 and HH2, where LL2 represents the second-layer low-frequency approximation subband; HL2, LH2 and HH2 represent the second-layer horizontal, vertical and diagonal detail subbands respectively;

[0024] The LL2 subband is decomposed into the third layer: LL3, HL3, LH3 and HH3, where LL3 represents the third layer low-frequency approximation subband, HL3, LH3 and HH3 represent the third layer horizontal, vertical and diagonal detail subbands respectively.

[0025] Furthermore, the adaptive thresholds are calculated respectively according to the subband categories, specifically including:

[0026] The high-frequency noise dominant sub-bands include: HH1 and HH2, and the high-frequency noise dominant threshold is set to T g , the calculation formula is:

[0027]

[0028] Where σ represents the noise standard deviation, N g represents the total number of pixels in the high-frequency noise-dominated subband, a g represents the median coefficient of high-frequency noise, M(·) represents the median function, |coeff g | represents the absolute value of the coefficient of the high-frequency noise-dominated subband after wavelet transform;

[0029] The horizontal lesion edge sub-bands include: HL1 and HL2, and the horizontal lesion edge threshold is set to T s , the calculation formula is:

[0030]

[0031] Among them, γ s represents the horizontal noise weight, a s Represents the coefficient of the median term of horizontal lesions, |coeff s | represents the absolute value of the coefficient of the horizontal lesion edge subband after wavelet transform;

[0032] The vertical lesion edge sub-bands include: LH1 and LH2, and the vertical lesion edge threshold is set to T c , the calculation formula is:

[0033]

[0034] Among them, γ c represents the vertical noise weight, a c represents the vertical spot median coefficient, |coeff c | represents the absolute value of the coefficient of the vertical spot edge subband after wavelet transform;

[0035] The low-frequency signal retention subbands include: LL3, HL3, LH3 and HH3. The low-frequency signal retention threshold is set to T d , the calculation formula is:

[0036]

[0037] Among them, γ d represents the low-frequency noise weight, a d represents the low-frequency noise coefficient, |coeff d | represents the absolute value of the coefficient of the low-frequency signal retention subband after wavelet transform.

[0038] Furthermore, the soft threshold function has the following specific formula:

[0039]

[0040] Where i represents different subband categories, i = [g, s, c, d], T i Indicates the threshold of different sub-band categories, D i represents the improved soft threshold function, c i It represents the high-frequency detail coefficient of different sub-band categories, sign(·) represents the positive and negative real number judgment function, and β represents the attenuation factor.

[0041] Furthermore, the plurality of pairs of denoised red bean leaf images are subjected to dual-channel fusion segmentation, specifically comprising the following steps:

[0042] Convert the RGB image to Lab color space, extract the brightness channel L and chrominance channel A, and the calculation formula is:

[0043]

[0044] Among them, R, G and B represent the three color channels of red, green and blue respectively;

[0045] Calculate the optimal threshold t for A channel and L channel based on OTSU algorithm A and t L ;

[0046] The luminance channel L and the chrominance channel A are fused to generate the rock red bean pest and disease area image. The calculation formula is:

[0047]

[0048] Among them, M(x,y) represents the retained pixel, A(x,y) represents the chromaticity of the pixel, and L(x,y) represents the brightness of the pixel.

[0049] Furthermore, the pest and disease diagnosis model specifically includes:

[0050] The input layer consists of 5 neurons, three convolutional layers, three pooling layers, a fully connected layer, and an output layer. The input layer uses Sigmoid as the activation function, and the convolutional layer and the pooling layer form a group. The output layer uses a Softmax classifier, and the output structure is a one-dimensional array of length 5.

[0051] The loss function calculation formula of the pest diagnosis model is:

[0052]

[0053] Among them, x represents the number of training samples, K represents the number of pest and disease categories, and p j,k represents the probability that sample j belongs to the kth pest label, y j,k represents the pest label of sample j, and λ represents the focusing factor.

[0054] Furthermore, the control plan is associated with a database of pest and disease types, control timing, and pesticide dosage, and a control plan is dynamically generated based on the probability classification of pests and diseases.

[0055] According to one aspect of the present invention, a storage medium is provided, wherein instructions are stored in the storage medium. When a computer reads the instructions, the computer executes any one of the above-mentioned methods for diagnosing and controlling diseases and pests of transplanted rock beans.

[0056] According to another aspect of the present invention, an electronic device is provided, comprising a processor and the above-mentioned storage medium, wherein the processor executes instructions in the storage medium.

[0057] Compared with the prior art, the present invention has the following beneficial effects:

[0058] 1. The present invention addresses the problem that traditional wavelet denoising methods are prone to loss of lesion details or residual noise due to fixed thresholds. The adaptive wavelet threshold technology achieves a balance between noise suppression and lesion feature retention by dynamically adjusting the threshold parameters.

[0059] 2. The present invention improves the soft threshold function of wavelet denoising, retaining weak spot signals while suppressing noise, avoiding the problem of traditional soft threshold completely losing weak edges.

[0060] 3. The three-layer wavelet decomposition of this invention provides a multi-resolution analysis framework for images of red bean pests and diseases by extracting subbands of different scales and directions. Combined with an adaptive thresholding strategy, it can accurately separate noise and lesion characteristics, laying the foundation for subsequent diagnosis and prevention.

[0061] 4. The present invention uses dual-channel threshold segmentation, which has strong anti-interference ability. The disease usually appears as a high value in the A channel and a low value in the L channel. The dual-channel constraint reduces mis-segmentation, and the complexity is comparable to that of the single-channel Otsu, which is suitable for real-time processing.

[0062] 5. The present invention increases the weight of difficult-to-classify samples (such as fuzzy spots) by setting a focusing factor in the loss function, focusing on leaf spots rather than healthy textures, and improving the recognition accuracy of local diseases. BRIEF DESCRIPTION OF THE DRAWINGS

[0063] Other features, objects and advantages of the present invention will become more apparent upon reading the detailed description of non-limiting embodiments made with reference to the following drawings:

[0064] Figure 1 A schematic diagram of a flow chart of an embodiment of the present invention;

[0065] Figure 2 This is a flow chart of wavelet denoising according to an embodiment of the present invention;

[0066] Figure 3 Schematic diagram of wavelet denoising decomposition according to an embodiment of the present invention;

[0067] Figure 4 This is a diagram of the pest diagnosis model architecture of an embodiment of the present invention. DETAILED DESCRIPTION

[0068] In order to make the objectives, technical solutions and advantages of the present invention more clear, the present invention is described in detail below with reference to the accompanying drawings and specific embodiments.

[0069] like Figure 1 As shown, the method for diagnosing and controlling diseases and insect pests of transplanted rock red beans includes the following steps:

[0070] Deploy video surveillance to collect images of rock red bean leaves;

[0071] Adaptive wavelet threshold denoising was performed on the collected images of Echinops lithophylla leaves;

[0072] The denoised red bean leaf image is subjected to dual-channel fusion segmentation to obtain the red bean disease and insect pest area image;

[0073] Construct a pest and disease diagnosis model based on CNN neural network;

[0074] The image of the pest and disease area of ​​rock red bean is input into the pest and disease diagnosis model and the pest and disease probability is output;

[0075] When the probability of pests and diseases exceeds the threshold, a prevention and control plan is generated.

[0076] like Figure 2 As shown, the collected transplanted rock red bean leaf image is subjected to adaptive wavelet threshold denoising, which specifically includes the following steps:

[0077] Perform three-layer wavelet decomposition on the collected leaf images of red bean:

[0078] The three-layer wavelet decomposition is divided into subbands and adaptive thresholds are calculated according to subband categories, wherein the subband categories include: high-frequency noise dominant category, horizontal lesion edge category, vertical lesion edge category, and low-frequency signal retention category;

[0079] Improve the soft threshold function to selectively suppress the high-frequency coefficients after wavelet decomposition;

[0080] The wavelet was reconstructed to generate the denoised image of the leaf of Echinops lithophylla.

[0081] like Figure 3 As shown, the three-layer wavelet decomposition of the collected leaf image of the rock red bean specifically includes:

[0082] The first layer is decomposed into: LL1, HL1, LH1 and HH1, where LL1 represents the low-frequency approximation subband; HL1, LH1 and HH1 represent the high-frequency detail subbands in the horizontal, vertical and diagonal directions respectively;

[0083] The LL1 subband is decomposed into the following two layers: LL2, HL2, LH2 and HH2, where LL2 represents the second-layer low-frequency approximation subband; HL2, LH2 and HH2 represent the second-layer horizontal, vertical and diagonal detail subbands respectively;

[0084] The LL2 subband is decomposed into the third layer: LL3, HL3, LH3 and HH3, where LL3 represents the third layer low-frequency approximation subband, HL3, LH3 and HH3 represent the third layer horizontal, vertical and diagonal detail subbands respectively.

[0085] The adaptive thresholds are calculated according to the sub-band categories, specifically including:

[0086] The high-frequency noise dominant sub-bands include: HH1 and HH2, and the high-frequency noise dominant threshold is set to T g , the calculation formula is:

[0087]

[0088] Where σ represents the noise standard deviation, N g represents the total number of pixels in the high-frequency noise-dominated subband, a g represents the median coefficient of high-frequency noise, M(·) represents the median function, |coeff g | represents the absolute value of the coefficient of the high-frequency noise-dominated subband after wavelet transform;

[0089] The horizontal lesion edge sub-bands include: HL1 and HL2, and the horizontal lesion edge threshold is set to T s , the calculation formula is:

[0090]

[0091] Among them, γ s represents the horizontal noise weight, a s Represents the coefficient of the median term of horizontal lesions, |coeff s | represents the absolute value of the coefficient of the horizontal lesion edge subband after wavelet transform;

[0092] The vertical lesion edge sub-bands include: LH1 and LH2, and the vertical lesion edge threshold is set to T c , the calculation formula is:

[0093]

[0094] Among them, γ c represents the vertical noise weight, a c represents the vertical spot median coefficient, |coeff c | represents the absolute value of the coefficient of the vertical spot edge subband after wavelet transform;

[0095] The low-frequency signal retention subbands include: LL3, HL3, LH3 and HH3. The low-frequency signal retention threshold is set to T d , the calculation formula is:

[0096]

[0097] Among them, γ d represents the low-frequency noise weight, a d represents the low-frequency noise coefficient, |coeff d | represents the absolute value of the coefficient of the low-frequency signal retention subband after wavelet transform.

[0098] As shown in Table 1, the sub-bands are divided into four categories according to the number of decomposition layers and directional characteristics.

[0099] Table 1

[0100]

[0101]

[0102] Table 2 shows the specific settings of the sub-band parameters.

[0103] Table 2

[0104]

[0105] The specific formula of the soft threshold function is:

[0106]

[0107] Where i represents different subband categories, i = [g, s, c, d], T i Indicates the threshold of different sub-band categories, D i represents the improved soft threshold function, c i It represents the high-frequency detail coefficient of different sub-band categories, sign(·) represents the positive and negative real number judgment function, and β represents the attenuation factor.

[0108] In early stage pest and disease images, the lesion signal is weak and the traditional soft threshold may over-suppress it.

[0109] By default, β=0.2, which retains weak spot signals while suppressing noise.

[0110] For example, if c = 0.8T, the traditional soft threshold outputs 0, while the improved soft threshold outputs 0.16T, avoiding completely missing weak edges.

[0111] As shown in Table 3, there is an experimental comparison between the traditional soft threshold and the improved soft threshold.

[0112] Table 3

[0113] method Lesion detection rate Residual noise Traditional soft thresholding 0.872 35.1dB Improved soft thresholding 0.935 34.8dB

[0114] The dual-channel fusion segmentation of the multiple pairs of denoised red bean leaf images specifically includes the following steps:

[0115] Convert the RGB image to Lab color space, extract the brightness channel L and chrominance channel A, and the calculation formula is:

[0116]

[0117] Among them, R, G and B represent the three color channels of red, green and blue respectively;

[0118] Calculate the optimal threshold t for A channel and L channel based on OTSU algorithm A and t L ;

[0119] The luminance channel L and the chrominance channel A are fused to generate the rock red bean pest and disease area image. The calculation formula is:

[0120]

[0121] Among them, M(x,y) represents the retained pixel, A(x,y) represents the chromaticity of the pixel, and L(x,y) represents the brightness of the pixel.

[0122] L channel: reflects brightness information and distinguishes the brightness difference between diseased spots and healthy tissues (for example, the root rot area has low brightness).

[0123] A channel: captures the difference in red and green shades (e.g., if the lesion is brown, the A value is higher).

[0124] Otsu's goal is to find the threshold t so that the inter-class variance of the two classes after segmentation is maximized. The specific formula is:

[0125] η(t)=ω0(t)·ω1(t)·[μ0(t)-μ1(t)] 2

[0126] Where ω0(t) and ω1(t) represent the percentage of pixels with values ​​less than or equal to t (background class) and the percentage of pixels with values ​​greater than t (foreground class), respectively. μ0(t) and μ1(t) represent the average grayscale value of the background class and the average grayscale value of the foreground class, respectively.

[0127] Traverse all possible thresholds t and find the t that maximizes η(t) as the optimal threshold.

[0128] like Figure 4 As shown, the pest and disease diagnosis model specifically includes:

[0129] The input layer consists of 5 neurons, three convolutional layers, three pooling layers, a fully connected layer, and an output layer. The input layer uses Sigmoid as the activation function, and the convolutional layer and the pooling layer form a group. The output layer uses a Softmax classifier, and the output structure is a one-dimensional array of length 5.

[0130] The loss function calculation formula of the pest diagnosis model is:

[0131]

[0132] Among them, x represents the number of training samples, K represents the number of pest and disease categories, and p j,k represents the probability that sample j belongs to the kth pest label, y j,k represents the pest label of sample j, and λ represents the focusing factor.

[0133] The basic structure of a CNN consists of an input layer, a convolutional layer, a pooling layer, a fully connected layer, and an output layer. The convolutional layer extracts image features, the pooling layer samples the features, the fully connected layer connects the extracted features, and the classifier outputs the results.

[0134] Image data is divided into training, validation, and test sets. The training set is a collection of data samples used for deep learning and training using a specified network and directly contributes to model parameter tuning. The network uses the training set to learn data features and generate a prediction model. The validation set is used to evaluate model performance and helps adjust parameters during training to avoid overfitting or underfitting. The test set is used to assess model accuracy and robustness. After the model is generated, it is used to test model performance using data that the model has not encountered before. During training, the data is divided into training, validation, and test sets in a ratio of 6:2:2.

[0135] The main types of diseases and pests of rock bean include the following:

[0136] 1. Standing tree decay. These are primarily caused by Ganoderma lucidum (Ganodermas p.) and Schizophyllum commne Fr. The former is highly infectious, primarily infecting the base of tree trunks, causing trunk decay and carrying significant potential damage. The latter is less pathogenic, primarily infecting wounds caused by mechanical damage, sunburn, frostbite, or bacterial infection. Once infected, it will gradually spread, causing the wound to persist for years or even gradually enlarge.

[0137] 2. Branch and trunk canker. The pathogen infects branches and trunks, causing the affected tissues to turn brown and necrotic. Symptoms manifest as long, longitudinally extending brown spots. This can cause branch dieback and trunk rot, and attract secondary pests such as bark beetles, accelerating the weakening of the tree.

[0138] 3. Wood borer. The larvae initially gather in clusters to eat the subcutaneous tissue (often hollowing it out), then disperse into the heartwood, often creating numerous holes inside the trunk, leading to the weakening or even death of the entire tree.

[0139] 4. Bark beetles. Adult bark beetles damage branches and trunks by feeding on them. Their symbiotic fungi are often highly infectious plant pathogens. The combined damage of the bark beetles and fungi can easily lead to weakening or even death of branches and trunks.

[0140] 5. Flower beetles. Many adult flower beetles were found on the tree trunks, but they mainly licked the sap that leaked from the damaged parts of the stems and did not cause any substantial harm to the trees themselves.

[0141] The control plan is associated with a database of pest and disease types, control timing, and pesticide dosage, and a control plan is dynamically generated based on the probability classification of pests and diseases.

[0142] The diagnosis result is a set of numbers, including the category code and probability of the pest and disease. If the probability is greater than 80%, the corresponding code is read from the database and the prevention and control suggestions are finally displayed to the user.

[0143] The computer-readable storage medium of this embodiment may be an internal storage unit of the terminal, such as a hard disk or memory of the terminal; the computer-readable storage medium of this embodiment may also be an external storage device of the terminal, such as a plug-in hard disk, a smart memory card, a secure digital card, a flash memory card, etc. equipped on the terminal; further, the computer-readable storage medium may also include both an internal storage unit of the terminal and an external storage device.

[0144] The computer-readable storage medium of this embodiment is used to store computer programs and other programs and data required by the terminal. The computer-readable storage medium can also be used to temporarily store data that has been output or is to be output.

[0145] Those skilled in the art will appreciate that all or part of the processes in the above-described method embodiments can be implemented by instructing related hardware through a computer program. The program can be stored in a computer-readable storage medium, and when executed, the program can include the processes in the above-described method embodiments. The storage medium can be a magnetic disk, an optical disk, a read-only memory (ROM), or a random access memory (RAM).

[0146] The examples described in the present invention are merely descriptions of the preferred embodiments of the present invention and are not intended to limit the concept and scope of the present invention. Without departing from the design concept of the present invention, various modifications and improvements made to the technical solutions of the present invention by engineers and technicians in this field should fall within the scope of protection of the present invention.

Claims

1. A method for diagnosing and controlling diseases and insect pests of transplanted rock red beans, characterized in that: The steps include: Deploy video surveillance to collect images of rock red bean leaves; Adaptive wavelet threshold denoising was performed on the collected images of Echinops lithophylla leaves; The denoised red bean leaf image is subjected to dual-channel fusion segmentation to obtain the red bean disease and insect pest area image; Construct a pest and disease diagnosis model based on CNN neural network; The image of the pest and disease area of ​​rock red bean is input into the pest and disease diagnosis model and the pest and disease probability is output; When the probability of pests and diseases exceeds the threshold, a prevention and control plan is generated.

2. The method according to claim 1, characterized in that The adaptive wavelet threshold denoising of the collected transplanted rock red bean leaf image specifically includes the following steps: Perform three-layer wavelet decomposition on the collected leaf images of red bean: The three-layer wavelet decomposition is divided into subbands and adaptive thresholds are calculated according to subband categories, wherein the subband categories include: high-frequency noise dominant category, horizontal lesion edge category, vertical lesion edge category, and low-frequency signal retention category; Improve the soft threshold function to selectively suppress the high-frequency coefficients after wavelet decomposition; The wavelet was reconstructed to generate the denoised image of the leaf of Echinops lithophylla.

3. The method according to claim 2, characterized in that The three-layer wavelet decomposition of the collected red bean leaf image specifically includes: The first layer is decomposed into: LL1, HL1, LH1 and HH1, where LL1 represents the low-frequency approximation subband; HL1, LH1 and HH1 represent the high-frequency detail subbands in the horizontal, vertical and diagonal directions respectively; The LL1 subband is decomposed into the following two layers: LL2, HL2, LH2 and HH2, where LL2 represents the second-layer low-frequency approximation subband; HL2, LH2 and HH2 represent the second-layer horizontal, vertical and diagonal detail subbands respectively; The LL2 subband is decomposed into the third layer: LL3, HL3, LH3 and HH3, where LL3 represents the third layer low-frequency approximation subband, HL3, LH3 and HH3 represent the third layer horizontal, vertical and diagonal detail subbands respectively.

4. The method according to claim 2, characterized in that The adaptive thresholds are calculated according to the sub-band categories, specifically including: The high-frequency noise dominant sub-bands include: HH1 and HH2, and the high-frequency noise dominant threshold is set to T g , the calculation formula is: Where σ represents the noise standard deviation, N g represents the total number of pixels in the high-frequency noise-dominated subband, a g represents the median coefficient of high-frequency noise, M(·) represents the median function, |coeff g | represents the absolute value of the coefficient of the high-frequency noise-dominated subband after wavelet transform; The horizontal lesion edge sub-bands include: HL1 and HL2, and the horizontal lesion edge threshold is set to T s , the calculation formula is: Among them, γ s represents the horizontal noise weight, a s Represents the coefficient of the median term of horizontal lesions, |coeff s | represents the absolute value of the coefficient of the horizontal lesion edge subband after wavelet transform; The vertical lesion edge sub-bands include: LH1 and LH2, and the vertical lesion edge threshold is set to T c , the calculation formula is: Among them, γ c represents the vertical noise weight, a c represents the vertical spot median coefficient, |coeff c | represents the absolute value of the coefficient of the vertical spot edge subband after wavelet transform; The low-frequency signal retention subbands include: LL3, HL3, LH3 and HH3. The low-frequency signal retention threshold is set to T d , the calculation formula is: Among them, γ d represents the low-frequency noise weight, a d represents the low-frequency noise coefficient, |coeff d | represents the absolute value of the coefficient of the low-frequency signal retention subband after wavelet transform.

5. The method according to claim 2, characterized in that The specific formula of the soft threshold function is: Where i represents different subband categories, i = [g, s, c, d], T i Indicates the threshold of different sub-band categories, D i represents the improved soft threshold function, c i It represents the high-frequency detail coefficient of different sub-band categories, sign(·) represents the positive and negative real number judgment function, and β represents the attenuation factor.

6. The method according to claim 1, characterized in that The dual-channel fusion segmentation of the multiple pairs of denoised red bean leaf images specifically includes the following steps: Convert the RGB image to Lab color space, extract the brightness channel L and chrominance channel A, and the calculation formula is: Among them, R, G and B represent the three color channels of red, green and blue respectively; Calculate the optimal threshold t for A channel and L channel based on OTSU algorithm A and t L ; The luminance channel L and the chrominance channel A are fused to generate the rock red bean pest and disease area image. The calculation formula is: Among them, M(x,y) represents the retained pixel, A(x,y) represents the chromaticity of the pixel, and L(x,y) represents the brightness of the pixel.

7. The method according to claim 1, characterized in that The pest and disease diagnosis model specifically includes: The input layer consists of 5 neurons, three convolutional layers, three pooling layers, a fully connected layer, and an output layer. The input layer uses Sigmoid as the activation function, and the convolutional layer and the pooling layer form a group. The output layer uses a Softmax classifier, and the output structure is a one-dimensional array of length 5. The loss function calculation formula of the pest diagnosis model is: Among them, x represents the number of training samples, K represents the number of pest and disease categories, and p j,k represents the probability that sample j belongs to the kth pest label, y j,k represents the pest label of sample j, and λ represents the focusing factor.

8. The method according to claim 1, characterized in that The control plan is associated with a database of pest and disease types, control timing, and pesticide dosage, and a control plan is dynamically generated based on the probability classification of pests and diseases.

9. A storage medium, characterized in that: The storage medium stores instructions, and when a computer reads the instructions, the computer is caused to execute the method for diagnosing and controlling diseases and pests of transplanted rock bean according to any one of claims 1 to 7.

10. An electronic device, characterized in that: The device comprises a processor and the storage medium according to claim 9, wherein the processor executes instructions in the storage medium.

Images