Tea round brown spot hazard grade detection method based on hyperspectral technology

Abstract

The invention provides a tea round brown spot hazard grade detection method based on a hyperspectral technology, and the method comprises the steps: obtaining the original hyperspectral data of tea leaves in a tea garden, the hyperspectral technology is high in resolution, and the hyperspectral reflectivity information of continuous wave bands of the tea leaves in the tea garden can be rapidly obtained. The original hyperspectral data are preprocessed, preprocessed data are obtained, and preprocessing comprises filtering and first-order derivation; the first-order derivation can amplify detail information and noise in the spectrum, so that the original hyperspectral data is firstly filtered, and then the filtered data is subjected to first-order derivation, thereby eliminating baseline drift, reducing interference caused by a background area, and enabling the preprocessed data to have a relatively high resolution. According to the method, the tea round brown spot detection model is constructed, the preprocessed data are input into the tea round brown spot detection model, hazard grade judgment is carried out, the tea round brown spot hazard grade is obtained, the tea round brown spot hazard grade can be effectively detected, and the method has high recognition accuracy.

Description

Technical Field

[0001] The invention relates to the technical field of tea tree disease detection, and in particular to a method for detecting the hazard level of tea leaf brown spot disease based on hyperspectral technology. Background Art

[0002] Tea leaf spot disease is one of the common leaf diseases of tea trees. When the disease occurs, the diseased leaves appear in pieces and are locally contagious. It mainly occurs in tea gardens with high humidity and fog. The probability of tea leaf spot disease is higher in early spring. The tea leaves that are infected appear purple-brown, and a large number of diseased leaves fall off. The tea made from diseased leaves has a high rate of breakage, and the soup color is dark and bitter. Tea leaf spot disease has a great impact on the quality of tea.

[0003] The traditional identification method of tea leaf red spot disease mainly relies on manual identification based on characteristic symptoms or through pathological examination, followed by the use of images to effectively identify crop diseases and estimate the condition. However, the above methods cannot effectively distinguish the onset period of tea leaf red spot disease.

[0004] Therefore, there is a need for a detection method that can detect tea round red spot disease of different disease damage levels and has a high recognition accuracy rate. Summary of the invention

[0005] In order to overcome the problems existing in the related art, the purpose of the present invention is to provide a method for detecting the hazard level of tea round brown spot disease based on hyperspectral technology. The method combines hyperspectral technology and convolutional neural network, and introduces a gated loop module in the convolutional neural network. It can effectively detect the hazard level of tea round brown spot disease and has a high recognition accuracy.

[0006] A method for detecting the damage level of tea round brown spot disease based on hyperspectral technology, comprising:

[0007] Acquire original hyperspectral data of tea leaves in a tea garden; wherein the tea leaves in the tea garden include tea leaves suffering from brown spot disease and healthy tea leaves;

[0008] Preprocessing the original hyperspectral data to obtain preprocessed data; the preprocessing includes filtering and first-order derivative;

[0009] A tea round brown spot disease detection model is constructed, and the preprocessed data is input into the tea round brown spot disease detection model to make a hazard level judgment to obtain the hazard level of the tea round brown spot disease; the tea round brown spot disease detection model is based on a convolutional neural network and a gated loop module.

[0010] In a preferred technical solution of the present invention, the preprocessing of the original hyperspectral data to obtain preprocessed data includes:

[0011] Performing a standard normal transformation on the original hyperspectral data to obtain transformed spectral data;

[0012] filtering the transformed spectral data to obtain filtered data;

[0013] Performing a first-order derivative on the filtered data to obtain preprocessed data.

[0014] In a preferred technical solution of the present invention, the step of performing a standard normal transformation on the original hyperspectral data to obtain transformed spectral data includes:

[0015] The original hyperspectral data is subjected to a standard normal transformation according to the following formula:

[0016]

[0017] Among them, x i is the average spectral reflectance of the i-th sample, x i,k is the spectral reflectance of the kth spectral band of the i-th sample; m is the total number of spectral bands of the i-th sample, X i,SNV is the spectral data of the i-th sample after transformation, 1≤i≤n, n is the total number of samples;

[0018] The spectral data of all transformed samples are combined to obtain transformed spectral data.

[0019] In a preferred technical solution of the present invention, filtering the transformed spectral data to obtain filtered data includes:

[0020] The transformed spectral data is subjected to SG filtering according to the following formula:

[0021]

[0022] Among them, a j is the jth polynomial coefficient, r is the size of the SG window, m1 is the reference coefficient, X i,SG is the spectral data after SG filtering of the i-th;

[0023] All spectral data after SG filtering are combined to obtain filtered data.

[0024] In a preferred technical solution of the present invention, performing a first-order derivative on the filtered data to obtain preprocessed data includes:

[0025] The first-order derivative of the filtered data is taken according to the following formula:

[0026]

[0027] in, is the wavelength λ i+1 Reflectance after standard normal transformation and SG filtering, is the wavelength λ i-1 Reflectivity after standard normal transformation and SG filtering, ΔX is the wavelength λ i-1 To wavelength λ i+1 The wavelength interval, X'(X i，SG ) is the preprocessed data.

[0028] In a preferred technical solution of the present invention, the method of obtaining the original hyperspectral data of tea leaves in the tea garden includes:

[0029] Obtaining the tea leaves infected with tea leaf brown spot disease and the healthy tea leaves;

[0030] A hyperspectral imager is used to collect spectral reflectance data of the tea leaves suffering from tea leaf brown spot disease and the healthy tea leaves to obtain original hyperspectral data of the tea leaves in the tea garden; wherein the spectral band range of the hyperspectral imager is 400-1000nm.

[0031] In a preferred technical solution of the present invention, the construction of a tea leaf brown spot disease detection model comprises:

[0032] The CNN-GRU neural network is used as the network architecture of the tea round brown spot disease detection model;

[0033] Using 5 one-dimensional convolutional layers as the feature extraction module of the tea round red spot disease detection model;

[0034] The input end of the feature extraction module is connected to the input module, and the output end of the feature extraction module is connected to the input end of the global average pooling layer.

[0035] In a preferred technical solution of the present invention, after connecting the output end of the feature extraction module with the input end of the global average pooling layer, it also includes:

[0036] Connecting the output of the global average pooling layer to the input of the gated recurrent module;

[0037] Connecting the output of the gated recurrent module to the input of the fully connected layer;

[0038] The output end of the fully connected layer is connected to the input end of the activation layer; the output end of the activation layer outputs the damage level of tea round red spot disease.

[0039] In a preferred technical solution of the present invention, the gated recurrent module includes two GRU layers and one dropout layer, wherein the output end of one GRU layer is connected to the input end of the dropout layer, and the output end of the dropout layer is connected to the input end of another GRU layer.

[0040] In a preferred technical solution of the present invention, the damage levels of the tea round red spot disease include healthy, mild, moderate and severe, wherein the number of lesions corresponding to healthy is 0, the number of lesions corresponding to mild is 1-30, the number of lesions corresponding to moderate is 31-60, and the number of lesions corresponding to severe is greater than 60.

[0041] The beneficial effects of the present invention are:

[0042] The method for detecting the hazard level of tea leaf red spot disease based on hyperspectral technology provided by the present invention comprises obtaining the original hyperspectral data of tea leaves in a tea garden; wherein the tea leaves in the tea garden include tea leaves suffering from tea leaf red spot disease and healthy tea leaves. The hyperspectral technology has high resolution and can quickly obtain the hyperspectral reflectance information of continuous bands of tea leaves in the tea garden. The original hyperspectral data is preprocessed to obtain preprocessed data, and the preprocessing includes filtering and first-order derivative. The first-order derivative can amplify the detail information in the spectrum, and at the same time amplify the noise, so when the spectrum is derivatized, the original hyperspectral data is first filtered, and then the filtered data is first derived, thereby eliminating the baseline drift and reducing the interference caused by the background area, so that the preprocessed data has a higher resolution. A tea leaf red spot disease detection model is constructed, and the preprocessed data is input into the tea leaf red spot disease detection model to judge the hazard level and obtain the tea leaf red spot disease hazard level. The tea leaf red spot disease detection model is based on a convolutional neural network and a gated loop module. The convolutional neural network can automatically learn the deep features of the preprocessed data and can be used to detect the hazard level of tea leaf red spot disease. The gated loop module is used to solve the problem of gradient vanishing in convolutional neural networks. Combining hyperspectral technology, convolutional neural networks and gated loop modules, the damage level of tea round red spot disease can be effectively detected with a high recognition accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] Figure 1 It is a flow chart of the method for detecting the hazard level of tea round brown spot disease based on hyperspectral technology of the present invention;

[0044] Figure 2 It is a flow chart of preprocessing raw hyperspectral data of the present invention;

[0045] Figure 3 It is a flow chart of constructing a tea round brown spot disease detection model of the present invention;

[0046] Figure 4 It is a composition diagram of the tea round brown spot disease detection model of the present invention. DETAILED DESCRIPTION

[0047] The preferred embodiments of the present invention will be described in more detail below with reference to the accompanying drawings. Although the preferred embodiments of the present invention are shown in the accompanying drawings, it should be understood that the present invention can be implemented in various forms and should not be limited by the embodiments described herein. On the contrary, these embodiments are provided to make the present invention more thorough and complete, and to fully convey the scope of the present invention to those skilled in the art.

[0048] Example 1

[0049] like Figure 1 As shown, this embodiment provides a method for detecting the hazard level of tea round brown spot disease based on hyperspectral technology, including:

[0050] S1: Acquire original hyperspectral data of tea leaves in a tea garden; wherein the tea leaves in the tea garden include tea leaves suffering from brown spot disease and healthy tea leaves.

[0051] S2: Preprocessing the original hyperspectral data to obtain preprocessed data; the preprocessing includes filtering and first-order derivative.

[0052] S3: Construct a tea round brown spot disease detection model, input the preprocessed data into the tea round brown spot disease detection model, make a hazard level judgment, and obtain the hazard level of the tea round brown spot disease; the tea round brown spot disease detection model is based on a convolutional neural network and a gated loop module.

[0053] Step S1 includes the following steps S11-S12:

[0054] S11: Obtain the tea leaves infected with tea leaf brown spot disease and the healthy tea leaves.

[0055] S12: using a hyperspectral imager to collect spectral reflectance data of the tea leaves infected with tea leaf brown spot disease and the healthy tea leaves, to obtain original hyperspectral data of the tea leaves in the tea garden; wherein the spectral band range of the hyperspectral imager is 400-1000nm.

[0056] In this embodiment, the HyperSIS-VNIR-QE hyperspectral imager is used to collect spectral reflectance, and a hyperspectral image is drawn based on the spectral reflectance. The hyperspectral image is converted into a grayscale image, and the grayscale image is corrected by a black and white plate. The image after the black and white plate correction is used as the original hyperspectral data, and the original hyperspectral data includes the ROI area corresponding to the leaves in a single tea sample.

[0057] The average spectral reflectance in the ROI area is calculated, and the original hyperspectral data is preprocessed according to the average spectral reflectance to obtain the preprocessed data. The preprocessing includes filtering and first-order derivative. Filtering can eliminate the noise in the ROI area, and the first-order derivative can eliminate the influence of baseline drift and reduce the background interference, thereby improving the resolution of the ROI area.

[0058] The preprocessed data after step S2 is divided into a training set, a validation set and a test set in a ratio of 3:1:1, and the training set, the validation set and the test set are used as input variables in turn, and the damage level of tea leaf brown spot disease is used as the output variable. In this embodiment, the preprocessed data includes ROI areas corresponding to leaves in 1188 tea samples.

[0059] The tea round red star disease detection model is based on a neural network architecture, and a gated loop module is introduced into the neural network architecture. The training set is input into the tea round red star disease detection model to train the tea round red star disease detection model. The validation set is input into the trained tea round red star disease detection model to adjust the parameters of the trained tea round red star disease detection model. The test set is input into the tea round red star disease detection model after the adjusted parameters to obtain the tea round red star disease hazard level corresponding to the test set. Based on the tea round red star disease hazard level corresponding to the test set and the actual tea round red star disease hazard level, the accuracy, precision, recall and f1 score are calculated, and the tea round red star disease detection model is evaluated based on the accuracy, precision, recall and f1 score.

[0060] The tea round red star disease hazard level detection method based on hyperspectral technology provided in this embodiment includes obtaining the original hyperspectral data of tea leaves in a tea garden; wherein the tea leaves in the tea garden include tea leaves suffering from tea round red star disease and healthy tea leaves. Hyperspectral technology has high resolution and can quickly obtain the hyperspectral reflectance information of continuous bands of tea leaves in the tea garden. The original hyperspectral data is preprocessed to obtain preprocessed data, and the preprocessing includes filtering and first-order derivative. The first-order derivative can amplify the detailed information in the spectrum, and at the same time amplify the noise, so when the spectrum is derivatized, the original hyperspectral data is first filtered, and then the filtered data is first derived, thereby eliminating the baseline drift and reducing the interference caused by the background area, so that the preprocessed data has a higher resolution. A tea round red star disease detection model is constructed, and the preprocessed data is input into the tea round red star disease detection model to judge the hazard level and obtain the tea round red star disease hazard level. The tea round red star disease detection model is based on a convolutional neural network and a gated loop module. The convolutional neural network can automatically learn the deep features of the preprocessed data and can be used to detect the tea round red star disease hazard level of tea leaves. The gated loop module is used to solve the problem of gradient vanishing in convolutional neural networks. Combining hyperspectral technology, convolutional neural networks and gated loop modules, the damage level of tea round red spot disease can be effectively detected with a high recognition accuracy.

[0061] Example 2

[0062] This embodiment provides a method for detecting the hazard level of tea round brown spot disease based on hyperspectral technology. This embodiment only describes the differences from Example 1, such as Figure 2As shown, the preprocessing of the original hyperspectral data to obtain preprocessed data includes:

[0063] S21: performing a standard normal transformation on the original hyperspectral data to obtain transformed spectral data.

[0064] S22: Filtering the transformed spectral data to obtain filtered data.

[0065] S23: performing first-order derivative on the filtered data to obtain preprocessed data.

[0066] The step of performing a standard normal transformation on the original hyperspectral data to obtain transformed spectral data includes:

[0067] The original hyperspectral data is subjected to a standard normal transformation according to the following formula:

[0068]

[0069] Among them, x i is the average spectral reflectance of the i-th sample, x i,k is the spectral reflectance of the kth spectral band of the i-th sample; m is the total number of spectral bands of the i-th sample, X i,SNV is the spectral data of the i-th sample after transformation, 1≤i≤n, n is the total number of samples;

[0070] The spectral data of all transformed samples are combined to obtain transformed spectral data.

[0071] Each sample is represented by a spectral curve at the spectral level. The spectral curve is divided into multiple spectral bands during collection, and each spectral band corresponds to a spectral reflectance. The original hyperspectral data is subjected to a standard normal transformation so that the transformed spectral data meets the characteristics of normal distribution. Normal distribution has the advantages of centralization, symmetry, stability and independence, and has a high degree of fit to real data.

[0072] The filtering of the transformed spectral data to obtain filtered data includes:

[0073] The transformed spectral data is subjected to SG filtering according to the following formula:

[0074]

[0075] Among them, a j is the jth polynomial coefficient, r is the size of the SG window, m1 is the reference coefficient, X i,SG is the spectral data after SG filtering of the i-th;

[0076] All spectral data after SG filtering are combined to obtain filtered data.

[0077] SG filtering has the following advantages: (1) SG filtering can effectively reduce noise while maintaining data trends and shapes. (2) SG filtering can flexibly adjust the order of the fitting polynomial to adapt to different types of data. (3) SG filtering has faster calculation speed and lower storage requirements.

[0078] The first-order derivative of the filtered data is performed to obtain preprocessed data, including:

[0079] The first-order derivative of the filtered data is taken according to the following formula:

[0080]

[0081] in, is the wavelength λ i+1 Reflectance after standard normal transformation and SG filtering, is the wavelength λ i-1 Reflectivity after standard normal transformation and SG filtering, ΔX is the wavelength λ i-1 To wavelength λ i+1 The wavelength interval, X'(X i，SG ) is the preprocessed data.

[0082] The purpose of the first-order derivative is to eliminate baseline drift, reduce the impact of background interference, and provide higher resolution. The reflectance change is the baseline drift caused by the interference of the spectral acquisition environment. The first-order derivative can amplify the detailed information in the spectrum, but it will also amplify the noise. Therefore, the transformed spectral data is first filtered to remove the noise in the transformed spectral data, and then the first-order derivative of the filtered data is performed to obtain the pre-processed data.

[0083] This embodiment preprocesses the original hyperspectral data to obtain preprocessed data, including performing a standard normal transformation on the original hyperspectral data to obtain transformed spectral data. The transformed spectral data is filtered to obtain filtered data. The filtered data is first-order differentiated to obtain preprocessed data. The standard normal transformation makes the transformed spectral data more consistent with the distribution characteristics of the real data, filtering can remove the noise of the transformed spectral data, and the first-order derivative can eliminate baseline drift, reduce the influence of background interference, and provide higher resolution.

[0084] Example 3

[0085] like Figure 3As shown, this embodiment provides a method for detecting the hazard level of tea round red spot disease based on hyperspectral technology. This embodiment only describes the differences from Example 1. The construction of a tea round red spot disease detection model includes:

[0086] S31: The CNN-GRU neural network is used as the network architecture of the tea round brown spot disease detection model.

[0087] S32: Using 5 one-dimensional convolutional layers as a feature extraction module of the tea leaf red spot disease detection model.

[0088] S33: Connecting the input end of the feature extraction module to the input module, and connecting the output end of the feature extraction module to the input end of the global average pooling layer.

[0089] After connecting the output end of the feature extraction module to the input end of the global average pooling layer, the method further includes:

[0090] S34: Connecting the output end of the global average pooling layer to the input end of the gated recurrent module.

[0091] S35: Connecting the output end of the gated recurrent module to the input end of the fully connected layer.

[0092] S36: Connecting the output end of the fully connected layer to the input end of the activation layer; the output end of the activation layer outputs the damage level of tea round red spot disease.

[0093] Figure 4 : is a composition diagram of a tea round red spot disease detection model, wherein the gated recurrent module includes two GRU layers and one dropout layer, wherein the output end of one GRU layer is connected to the input end of the dropout layer, and the output end of the dropout layer is connected to the input end of another GRU layer. Figure 4 CNN in the figure represents convolutional neural network, Fully connected layer represents the fully connected layer, Softmax represents the normalized exponential function, Conv.Block1 represents the first convolutional layer, and Dropout represents the random dropout layer.

[0094] The CNN-GRU neural network of this embodiment has the following characteristics:

[0095] (1) The input spectral data size is 256*1, that is, 256 bands and 1 channel.

[0096] (2) Five one-dimensional convolutional layers are used to extract feature values. The number of convolution kernels is 16, the size is 1*5, and the stride is 2. The convolutional layer is used to extract the feature of the spectrum and mine the correlation between the data.

[0097] (3) Adding a global average pooling layer compresses the number of parameters, reduces model complexity, and improves generalization capabilities.

[0098] (4) Add the Dropout layer to randomly select neurons in the network to prevent overfitting.

[0099] (5) The GRU layer is used to process the reduced-dimensional data and learn the dependency relationship between different bands of spectral data.

[0100] (6) The fully connected layer flattens the data into a one-dimensional structure, and the Softmax activation function matches each feature data to the feature category with the highest probability.

[0101] Table 1 Classification criteria for tea leaf red spot disease

[0102] standard grade Number healthy 0 285 Mild (number of lesions 1-30) 1 289 Moderate (31-60 spots) 2 293 Severe (number of lesions>60) 3 321

[0103] As shown in Table 1, the damage levels of the tea round red spot disease in this embodiment include healthy, mild, moderate and severe, among which the number of lesions corresponding to healthy is 0, the number of lesions corresponding to mild is 1-30, the number of lesions corresponding to moderate is 31-60, and the number of lesions corresponding to severe is greater than 60.

[0104] In this embodiment, since the chlorophyll content of the leaves suffering from tea round red spot disease is more sensitive to the damage level of tea round red spot disease, chlorophyll absorbs most of the incident energy in two spectral bands with central wavelengths of 450nm, i.e., blue light and 650nm, i.e., red light. Between these two chlorophyll absorption bands, due to the small absorption effect, a reflection peak is formed near 540nm, i.e., green light. The gated loop module includes multiple GRU layers, i.e., gated loop layers. When the band gap is large, the gated loop module can better capture the dependency relationship between different bands. The GRU layer is introduced into the convolutional neural network to establish a tea round red spot disease detection model that can output different tea round red spot disease damage levels.

[0105] Example 4

[0106] A computer device is also provided in an embodiment of the present application, which may be a server, wherein the computer device includes a processor, a memory, a network interface and a database connected via a system bus. The processor of the computer design is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The network interface of the computer device is used to communicate with an external terminal via a network connection.

[0107] This embodiment also provides a computer-readable storage medium on which a computer program is stored. When the computer program is executed by a processor, a method for detecting the hazard level of tea round red spot disease based on hyperspectral technology is implemented. It can be understood that the computer-readable storage medium in this embodiment can be a volatile readable storage medium or a non-volatile readable storage medium.

[0108] It should be noted that, in this article, the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, device, article or method including a series of elements includes not only those elements, but also includes other elements not explicitly listed, or also includes elements inherent to such process, device, article or method. In the absence of further restrictions, an element defined by the sentence "includes a ..." does not exclude the presence of other identical elements in the process, device, article or method including the element.

[0109] The above description is only a preferred embodiment of the present application, and does not limit the patent scope of the present application. Any equivalent structure or equivalent process transformation made using the contents of the present application specification and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present application.

Claims

1. A method for detecting the hazard level of tea leaf brown spot disease based on hyperspectral technology, characterized in that: include: Acquire original hyperspectral data of tea leaves in a tea garden; wherein the tea leaves in the tea garden include tea leaves suffering from brown spot disease and healthy tea leaves; Preprocessing the original hyperspectral data to obtain preprocessed data; the preprocessing includes filtering and first-order derivative; A tea round brown spot disease detection model is constructed, and the preprocessed data is input into the tea round brown spot disease detection model to make a hazard level judgment to obtain the hazard level of the tea round brown spot disease; the tea round brown spot disease detection model is based on a convolutional neural network and a gated loop module.

2. The method for detecting the hazard level of tea round brown spot disease based on hyperspectral technology according to claim 1, characterized in that: The preprocessing of the original hyperspectral data to obtain preprocessed data includes: Performing a standard normal transformation on the original hyperspectral data to obtain transformed spectral data; filtering the transformed spectral data to obtain filtered data; Performing a first-order derivative on the filtered data to obtain preprocessed data.

3. The method for detecting the hazard level of tea round brown spot disease based on hyperspectral technology according to claim 2 is characterized in that: The step of performing a standard normal transformation on the original hyperspectral data to obtain transformed spectral data includes: The original hyperspectral data is subjected to a standard normal transformation according to the following formula: Among them, x i is the average spectral reflectance of the i-th sample, x i,k is the spectral reflectance of the kth spectral band of the i-th sample; m is the total number of spectral bands of the i-th sample, X i,SNV is the spectral data of the i-th sample after transformation, 1≤i≤n, n is the total number of samples; The spectral data of all transformed samples are combined to obtain transformed spectral data.

4. The method for detecting the hazard level of tea round brown spot disease based on hyperspectral technology according to claim 3 is characterized in that: The filtering of the transformed spectral data to obtain filtered data includes: The transformed spectral data is subjected to SG filtering according to the following formula: Among them, a j is the jth polynomial coefficient, r is the size of the SG window, m1 is the reference coefficient, X i,SG is the spectral data after SG filtering of the i-th; All spectral data after SG filtering are combined to obtain filtered data.

5. The method for detecting the hazard level of tea round brown spot disease based on hyperspectral technology according to claim 2 is characterized in that: The first-order derivative of the filtered data is performed to obtain preprocessed data, including: The first-order derivative of the filtered data is taken according to the following formula: in, is the wavelength λ i+1 Reflectance after standard normal transformation and SG filtering, is the wavelength λ i-1 Reflectivity after standard normal transformation and SG filtering, ΔX is the wavelength λ i-1 To wavelength λ i+1 The wavelength interval, X'(X i，SG ) is the preprocessed data.

6. The method for detecting the hazard level of tea round brown spot disease based on hyperspectral technology according to claim 1, characterized in that: The method of obtaining the original hyperspectral data of tea leaves in the tea garden includes: Obtaining the tea leaves infected with tea leaf brown spot disease and the healthy tea leaves; A hyperspectral imager is used to collect spectral reflectance data of the tea leaves suffering from tea leaf brown spot disease and the healthy tea leaves to obtain original hyperspectral data of the tea leaves in the tea garden; wherein the spectral band range of the hyperspectral imager is 400-1000nm.

7. The method for detecting the hazard level of tea round brown spot disease based on hyperspectral technology according to claim 1, characterized in that: The method of constructing a tea leaf brown spot disease detection model comprises: The CNN-GRU neural network is used as the network architecture of the tea round brown spot disease detection model; Using 5 one-dimensional convolutional layers as the feature extraction module of the tea round red spot disease detection model; The input end of the feature extraction module is connected to the input module, and the output end of the feature extraction module is connected to the input end of the global average pooling layer.

8. The method for detecting the hazard level of tea round brown spot disease based on hyperspectral technology according to claim 7 is characterized in that: After connecting the output end of the feature extraction module to the input end of the global average pooling layer, the method further includes: Connecting the output of the global average pooling layer to the input of the gated recurrent module; Connecting the output of the gated recurrent module to the input of the fully connected layer; The output end of the fully connected layer is connected to the input end of the activation layer; the output end of the activation layer outputs the damage level of tea round red spot disease.

9. The method for detecting the hazard level of tea round brown spot disease based on hyperspectral technology according to claim 8, characterized in that: The gated recurrent module includes two GRU layers and a dropout layer, wherein the output end of one GRU layer is connected to the input end of the dropout layer, and the output end of the dropout layer is connected to the input end of another GRU layer.

10. The method for detecting the hazard level of tea round brown spot disease based on hyperspectral technology according to claim 1, characterized in that: The damage levels of tea round red spot disease include healthy, mild, moderate and severe, among which healthy corresponds to 0 lesions, mild corresponds to 1-30 lesions, moderate corresponds to 31-60 lesions, and severe corresponds to more than 60 lesions.

Images