Potato seed potato disease identification method based on high-resolution spectral features
By using high-resolution spectral characterization methods to segment sprouted and non-sprouting regions, calculating wavelength variation evaluation coefficients and salience coefficients, and constructing disease indices, the high cost and low efficiency of existing detection methods are solved, enabling rapid and accurate identification of potato seed tuber diseases.
Patent Information
- Application Number
- CN202311797615.7
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2023-12-26
- Publication Date
- 2024-11-15
- Estimated Expiration
- 2043-12-26
AI Technical Summary
Existing methods for detecting diseases in potato seed tubers are costly, inefficient, and prone to false positives, especially as they struggle to effectively remove interference from sprouting areas and redundant pixels.
A high-resolution spectral feature method was adopted, and the sprouted and non-sprouting regions were segmented by a semantic segmentation neural network. The wavelength variation evaluation coefficient and salience coefficient were calculated, and the wavelength pixel selection matrix and eigenvalue vector were constructed. Combined with the characteristics of normal potato seed tubers, the disease index was calculated for identification.
It improves detection accuracy and efficiency, reduces computational complexity, minimizes interference from redundant pixels, and enables rapid and accurate disease detection.
Smart Images

Figure CN117456370B_ABST
Abstract
Description
Technical Field
[0001] The present application relates to the technical field of spectral analysis, and in particular to a method for identifying potato seed potato diseases based on high-resolution spectral features. Background Art
[0002] Seed potatoes refer to seeds used to plant potatoes for the next season. The quality of these seeds directly affects the yield of potatoes for the next season, so it is necessary to inspect the seeds for diseases. The most common diseases of seed potatoes are late blight, early blight, and silver spot. When diseased seed potatoes are used for planting, it is easy to cause seedlings to die, resulting in economic losses. At present, the most accurate disease detection method for seed potatoes is mainly through pathogen detection. Although the accuracy of pathogen detection is high, the cost is high and the detection efficiency is low. In addition, when manually detecting diseases of seed potatoes, experienced technicians are needed to identify them. Although the cost is slightly lower, some diseases are hidden under the seed potato skin in the early stage, and direct identification by appearance is prone to false detection.
[0003] High-resolution spectral detection method refers to a method of measuring the interaction between matter and electromagnetic waves of different wavelengths by analyzing its hyperspectral data. By measuring the reflection of electromagnetic waves of different wavelengths by the matter, information about the composition, structure, ingredients and properties of the matter can be obtained. Using hyperspectral technology to detect diseases in seed potatoes is a fast, non-destructive and non-contact detection technology.
[0004] Since the surface of the seed potato includes a sprouting area and a non-sprouting area, in which the sprouting area is a hyperspectral area, which increases the difficulty of disease detection, the present invention adopts a spectral angle matching algorithm to detect whether the potato seed potato has disease. The hyperspectral data contains a large number of pixel sequences, and there is redundant information between different pixels, which can easily cause false detection of the results. Therefore, how to select effective features from a large number of pixel sequences for spectral angle matching is the focus of the present invention.
[0005] In summary, the present invention proposes a method for identifying potato seed potato diseases based on high-resolution spectral features, collects hyperspectral data of potato seed potatoes, eliminates the interference of germination area pixels and redundant pixels on disease area pixels in turn, and constructs a potato seed potato disease index based on the characteristics of normal potato seed potatoes to complete potato seed potato disease identification. Summary of the invention
[0006] In order to solve the above technical problems, the present invention provides a method for identifying potato seed potato diseases based on high-resolution spectral features to solve the existing problems.
[0007] The potato seed potato disease identification method based on high-resolution spectral characteristics of the present invention adopts the following technical scheme:
[0008] An embodiment of the present invention provides a method for identifying potato seed potato diseases based on high-resolution spectral features, the method comprising the following steps:
[0009] Obtain hyperspectral data of potato seed potatoes;
[0010] The hyperspectral data is synthesized into true color to obtain a pseudo color image; the pseudo color image is segmented into germination area and non-germination area by using semantic segmentation neural network; the non-germination area matrix is obtained based on the reflection intensity value of each pixel in the non-germination area in each band; the wavelength variation evaluation coefficient of each pixel is obtained according to the reflection intensity value of each pixel in the non-germination area matrix; the wavelength variation matrix of the non-germination area is obtained according to the wavelength variation evaluation coefficient of each pixel in the non-germination area matrix; the highlight coefficient of each row vector of the wavelength variation matrix of the non-germination area is calculated, and the highlight coefficients of all row vectors are arranged from large to small to construct a highlight wavelength selection sequence;
[0011] Adjust the non-germination area matrix and the non-germination area wavelength variation matrix according to the highlighted wavelength selection sequence; update the non-germination area wavelength variation matrix according to the wavelength variation evaluation coefficient of each pixel in the non-germination area wavelength variation matrix; obtain the wavelength pixel selection matrix according to the Hadamard product of the non-germination area matrix and the non-germination area wavelength variation matrix; obtain the eigenvalue vector of potato seed potatoes according to the wavelength pixel selection matrix; obtain the wavelength pixel selection matrix and eigenvalue vector of normal potato seed potatoes;
[0012] A potato seed potato disease index is obtained according to the wavelength pixel selection matrix and eigenvalue vector of potato seed potatoes and normal potato seed potatoes; and the disease condition of potato seed potatoes is identified according to the potato seed potato disease index.
[0013] Preferably, the non-germination area matrix is obtained based on the reflection intensity value of each pixel in the non-germination area in each band, including:
[0014] For each pixel in the non-germination area, the reflection intensity value of the pixel in each band is taken as the pixel sequence of the pixel;
[0015] The pixel sequence of each pixel in the non-germination area is used as a row vector to form a non-germination area matrix.
[0016] Preferably, obtaining the wavelength variation evaluation coefficient of each pixel according to the reflection intensity value of each pixel in the non-germination area matrix includes:
[0017] For each pixel in the matrix of the non-germination area, calculate the mean and standard deviation of the reflection intensity of all other pixels in the row vector where the pixel is located;
[0018] The absolute value of the difference between the reflection intensity value of the pixel and the mean value of the reflection intensity is calculated, and the ratio of the absolute value of the difference to the standard deviation of the reflection intensity is used as the wavelength variation assessment coefficient of the pixel.
[0019] Preferably, obtaining the wavelength variation matrix of the non-germination area according to the wavelength variation evaluation coefficient of each pixel of the non-germination area matrix includes:
[0020] The reflection intensity value of each pixel in the non-germination area matrix was replaced with the wavelength variation evaluation coefficient of each pixel, and the obtained matrix was used as the wavelength variation matrix of the non-germination area.
[0021] Preferably, the calculation of the highlight coefficient of each row vector of the wavelength variation matrix of the non-germination area includes:
[0022] For each row vector of the wavelength variation matrix in the non-germination area, the mean of the wavelength variation assessment coefficients of all band pixels in the row vector is taken as the highlight coefficient of the row vector.
[0023] Preferably, the adjusting of the non-germination area matrix and the non-germination area wavelength variation matrix according to the highlighted wavelength selection sequence comprises:
[0024] The first m% pixels in the highlight wavelength selection sequence are used as the highlight pixels of the non-germination area, where m is the preset selection coefficient;
[0025] The highlighted pixels in the non-germination area are screened out from the non-germination area matrix and the non-germination area wavelength variation matrix as the adjusted non-germination area matrix and the non-germination area wavelength variation matrix.
[0026] Preferably, the updating of the wavelength variation matrix of the non-germination area according to the wavelength variation evaluation coefficient of each pixel of the wavelength variation matrix of the non-germination area includes:
[0027] For each pixel in the wavelength variation matrix of the non-germination area, when the wavelength variation assessment coefficient of the pixel is less than the screening threshold, the wavelength variation assessment coefficient of the pixel is set to zero; when the wavelength variation assessment coefficient of the pixel is greater than or equal to the screening threshold, the wavelength variation assessment coefficient of the pixel remains unchanged.
[0028] Preferably, obtaining the eigenvalue vector of the seed potato according to the wavelength pixel selection matrix includes:
[0029] Singular value decomposition is used to calculate the eigenvalues of the wavelength pixel selection matrix, and all eigenvalues are combined into the eigenvalue vector of potato seed potatoes.
[0030] Preferably, the potato seed potato disease index is obtained according to the wavelength pixel selection matrix and eigenvalue vector of the potato seed potato and the normal potato seed potato, and the expression is:
[0031] ;
[0032] In the formula, represents the potato seed potato disease index, represents the wavelength pixel selection matrix for potato seed potatoes, Represents the wavelength pixel selection matrix of normal potato seed potatoes, and Respectively represent the number of pixel sequences in the wavelength pixel selection matrix of seed potato and normal potato, represents the F-norm of the matrix, and represent the eigenvalue vectors of seed potato and normal potato respectively, Indicates the modulus of the calculated vector, Indicates the number of bands.
[0033] Preferably, the step of identifying the disease condition of seed potatoes according to the seed potato disease index comprises:
[0034] When the potato seed potato disease index is less than or equal to the first threshold, the potato seed potatoes are identified as healthy seed potatoes;
[0035] When the potato seed potato disease index is greater than the first threshold value and less than or equal to the second threshold value, the potato seed potatoes are identified as qualified seed potatoes;
[0036] When the potato seed potato disease index is greater than the second threshold, the potato seed potatoes are identified as diseased seed potatoes, wherein the first threshold is less than the second threshold.
[0037] The present invention has at least the following beneficial effects:
[0038] The present invention divides seed potatoes into a germination area and a non-germination area, eliminates the influence of the germination area through semantic segmentation, constructs a non-germination area matrix, and excludes the identification of potato seed potato diseases by the germination area, so that the result is more accurate; the present invention calculates the wavelength variation evaluation coefficient for each pixel in the non-germination area matrix, constructs the non-germination area wavelength variation matrix, so that the highlight coefficient of each pixel in the non-germination area can be calculated, and a highlight pixel selection sequence is obtained. By screening out the highlight pixels, the pixels in the diseased area are screened out from the pixels in the normal area, and redundant pixels are eliminated, thereby reducing the time complexity of the algorithm;
[0039] The wavelength pixel selection matrix is obtained by combining the wavelength variation matrix of the non-germination area and the matrix of the non-germination area, and the corresponding eigenvalue vector is obtained, which can reflect the characteristics of normal seed potato. The difference between the wavelength pixel selection matrix and the eigenvalue vector of the normal seed potato is calculated to obtain the seed potato disease index, and the disease degree of the seed potato is indirectly reflected by the difference between the normal seed potato and the seed potato. The present invention eliminates the interference of the germinated area pixels and redundant pixels, optimizes the matrix by technical means, obtains the corresponding eigenvalue vector, and finally constructs the seed potato disease index, avoids the calculation consumption caused by calculating a large number of redundant pixel sequences, improves the detection efficiency, and reduces the interference of pixel sequence noise by extracting features of the diseased area, thereby improving the detection accuracy. BRIEF DESCRIPTION OF THE DRAWINGS
[0040] In order to more clearly illustrate the technical solutions and advantages in the embodiments of the present invention or the prior art, the drawings required for use in the embodiments or the prior art descriptions are briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.
[0041] Figure 1 A flow chart of a method for identifying potato seed potato diseases based on high-resolution spectral features provided by the present invention. DETAILED DESCRIPTION
[0042] In order to further explain the technical means and effects adopted by the present invention to achieve the predetermined invention purpose, the potato seed potato disease identification method based on high-resolution spectral characteristics proposed by the present invention, its specific implementation method, structure, characteristics and effects are described in detail as follows in conjunction with the accompanying drawings and preferred embodiments. In the following description, different "one embodiment" or "another embodiment" does not necessarily refer to the same embodiment. In addition, specific features, structures or characteristics in one or more embodiments may be combined in any suitable form.
[0043] Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs.
[0044] The specific scheme of the potato seed potato disease identification method based on high-resolution spectral characteristics provided by the present invention is described in detail below with reference to the accompanying drawings.
[0045] An embodiment of the present invention provides a method for identifying potato seed potato diseases based on high-resolution spectral features.
[0046] Specifically, the following potato seed potato disease identification method based on high-resolution spectral features is provided, please refer to Figure 1, the method comprises the following steps:
[0047] Step S001, obtaining hyperspectral data of potato seed potatoes.
[0048] This embodiment proposes a potato seed potato disease identification method based on high-resolution spectral features through spectral analysis technology. In large potato farms, it is impossible to use traditional potato seed potato pathogen detection and manual detection methods. On the one hand, the detection cost is too high, and on the other hand, the detection efficiency is too low. Therefore, a hyperspectral camera is used for potato seed potato disease detection.
[0049] In order to obtain the hyperspectral data of potato seed potatoes, the Hyperion 3000 hyperspectral camera was used to obtain the hyperspectral data of potato seed potatoes on the production line. Since the acquisition of hyperspectral data is easily affected by the environment and equipment accuracy, the mean filtering method was used to filter the acquired hyperspectral data to reduce the impact of noise on the hyperspectral data.
[0050] In addition, since the obtained seed potato hyperspectral data contains a lot of redundant information, which is not conducive to improving the detection efficiency, the principal component analysis (PCA) algorithm is used to reduce the dimension of the seed potato hyperspectral data, reducing the original 250 bands to 20 bands. The principal component analysis (PCA) algorithm is a well-known technology and will not be described in detail in this embodiment.
[0051] The seed potato hyperspectral data after dimensionality reduction still has multiple channels, each channel represents the reflection intensity information of the seed potato in the corresponding band. The filtered seed potato hyperspectral data is recorded as , ,in Represents the hyperspectral data of potato seed potatoes, Indicates the quality of seed potatoes for the The reflection intensity information of each band, Indicates the number of bands; , Indicates Band The reflection intensity value of each pixel, Indicates The number of pixels in a band.
[0052] At this point, the hyperspectral data of potato seed potatoes can be obtained according to the above method, which is convenient for further analysis of potato seed potato disease identification.
[0053] Step S002, using the hyperspectral data of seed potato to eliminate the interference of germination area pixels and redundant pixels on disease area pixels, and constructing a seed potato disease index by taking normal potatoes as a reference.
[0054] If there is a disease on the seed potato, the disease area is only a small amount randomly distributed on the surface of the seed potato. However, there is a sprouting area on the surface of the seed potato, which will interfere with the disease detection of the seed potato and reduce the detection accuracy. The influence of the sprouting area needs to be eliminated.
[0055] By synthesizing the hyperspectral data of potato seed potatoes in true color, the multi-band hyperspectral data is fitted into a pseudo-color image, in which the pixel positions corresponding to each band in the hyperspectral data remain unchanged. Then, the G-FRNet semantic segmentation neural network model is used to perform 2-class region extraction on the pseudo-color image synthesized by hyperspectral true color, and the pseudo-color image is segmented into germination area and non-germination area. The pixels in the germination area are eliminated, and only the influence of the pixels in the non-germination area is analyzed.
[0056] In the corresponding hyperspectral data, the germination pixels are excluded and only the pixels in the non-germination area are retained to construct the non-germination area matrix ,in represents the number of pixels in the non-germination area, Indicates the number of bands, set the experience value It should be noted that in the non-germination area matrix, the number of pixels is much larger than the number of channels of the potato seed potato hyperspectral data, that is, .
[0057] In the non-germination area matrix, each row vector represents a pixel sequence, and each column vector represents the reflection intensity value of a band of potato seed potatoes at each pixel position. In order to facilitate understanding, the non-germination area matrix is described in detail:
[0058] ;
[0059] In the formula, Indicates size The non-germination area matrix, Represents the non-germination area matrix row vectors, Represents the non-germination area matrix The row vector is The reflection intensity value in each band, represents the number of pixels in the non-germination area, Indicates the number of bands.
[0060] In the non-germination area matrix, each element in the pixel sequence corresponding to each row vector represents the reflection intensity value of the pixel in different bands. Therefore, the wavelength variation assessment coefficient of the pixel is calculated for the reflection intensity value of the pixel in the jth band of the i-th pixel sequence.
[0061] ;
[0062] In the formula, Represents the non-germination area matrix Line The wavelength variation evaluation coefficient of the column pixels, Represents the non-germination area matrix Line The reflection intensity value of the column pixel, and Represents the non-germination region matrix Remove the first The mean and standard deviation of the reflection intensity of the pixels in the column.
[0063] It should be noted that It represents the wavelength variation evaluation coefficient of the corresponding position in the non-germination area matrix. The purpose of the formula is to amplify the most representative wavelength reflection intensity value in the pixel sequence. For example, in the pixel sequence, the reflection intensity values of most bands are relatively uniform, while the reflection intensity values of some bands are large. If the current band is a stable band, the mean and variance will not change much after removing the band, and the difference between the current band pixel and the mean is small, that is, The smaller it is; if the reflection intensity of the current band is large, the mean and variance of other pixels will decrease after the pixels of this band are removed, so the reflection intensity value of the current mutation band will be amplified, that is, The larger it is, the more prominent the pixel is, and the more abnormal the pixel is.
[0064] The wavelength variation evaluation coefficient is calculated for each pixel in the non-germination area matrix, and finally the wavelength variation matrix of the non-germination area with the same dimension as the non-germination area matrix can be obtained. .
[0065] Since the wavelength variation evaluation coefficients of each pixel in different row vectors in the wavelength variation matrix of the non-germination area are inconsistent, and the wavelength variation matrix of the non-germination area contains a large number of redundant pixel sequences, that is, the normal part of potato seed potatoes, the wavelength variation matrix of the non-germination area is used to evaluate the wavelength variation of each pixel in the non-germination area. Each row vector of constructs its highlight coefficient, and the highlight indicators of each row are combined into a highlight pixel selection sequence.
[0066] ;
[0067] In the formula, The wavelength variation matrix of the non-germination area is represented by The highlight coefficient of the row corresponds to the highlight pixel selection sequence elements, Indicates the number of bands, The wavelength variation matrix of the non-germination area is represented by Line The wavelength variation estimation coefficient of the column pixels.
[0068] It should be noted that It is a measure of the prominence of the pixel sequence of each row vector in the wavelength variation matrix of the non-germination area corresponding to each pixel in the non-germination area, and evaluates the prominence of each pixel in the non-germination area in all bands. The larger the wavelength variation evaluation coefficient of each pixel in each row vector, the larger the prominence coefficient of the row vector, that is, the more abnormal the pixel in the non-germination matrix is, the more prominent the pixel in the non-germination area is.
[0069] Since there are a large number of redundant pixels in the non-germination area, it is not conducive to subsequent calculations. After traversing all the row vectors of the wavelength variation matrix of the non-germination area, the pixels in the non-germination area are arranged in order from large to small according to the corresponding highlight coefficients to obtain the highlight pixel selection sequence.
[0070] The first 40% of pixels in the selection sequence of prominent pixels are selected as the prominent pixels in the non-germination area. For the prominent pixels, the number of prominent pixels in the non-germination area is set to , at this time the non-germination area matrix becomes , the wavelength variation matrix of the non-germination area becomes .
[0071] Wavelength variation matrix for non-germination areas The wavelength variation evaluation coefficient of each pixel in the image is used to set a screening threshold. , the pixels of each band that are less than the screening threshold are considered to be redundant pixels, which have no representative significance. The wavelength variation evaluation coefficient of the corresponding position is set to zero, and the wavelength variation evaluation coefficient of the pixels greater than or equal to the screening threshold is retained.
[0072] The wavelength variation matrix of the non-germination area and the non-germination area matrix obtained after screening are both matrices of the same dimension, and the wavelength pixel selection matrix is obtained by adopting the Hadamard product, that is, multiplying the corresponding elements in the two matrices. Among them, the Hadamard product is a well-known technology and will not be repeated in this embodiment.
[0073] ;
[0074] In the formula, The size is The wavelength pixel selection matrix, The size is The non-germination area matrix, The size is The wavelength variation matrix of the non-germination area, is the Hadamard product symbol, which means multiplying corresponding elements of matrices of the same dimension. , , , Both represent the number of bands. , Both represent the number of highlighted pixels in the non-germination area.
[0075] It should be noted that in the Hadamard product calculation process, when the pixel corresponds to a normal pixel of potato seed potatoes, the wavelength variation assessment coefficient of the pixel is zero, that is, the corresponding position of the obtained wavelength pixel selection matrix is zero; when the pixel is an abnormal pixel of potato, the wavelength variation assessment coefficient of the pixel is large, so the corresponding wavelength variation assessment coefficient will amplify the reflection intensity value of the pixel, that is, the reflection intensity value of the pixel at the corresponding position of the obtained wavelength pixel selection matrix is large, making the pixel more prominent.
[0076] The obtained wavelength pixel selection matrix By performing singular value decomposition, the eigenvalues of the wavelength pixel selection matrix can be obtained. Since the number of pixel sequences in the non-germination area is much larger than the number of bands, The number of eigenvalues depends only on the number of channels. So, according to the matrix of The number of bands can be obtained eigenvalues, together forming the eigenvalue vector, namely .
[0077] In order to compare the disease degree of seed potatoes, a normal seed potato needs to be selected as a reference standard, and the hyperspectral data of the normal seed potato is analyzed according to the same analysis method of the seed potato in this embodiment to construct the wavelength pixel selection matrix of the normal seed potato. ,in, represents the number of pixel sequences in the wavelength pixel selection matrix of normal potatoes, and The singular value decomposition algorithm is also used to obtain the eigenvalue vector of normal potato seed potatoes .
[0078] The wavelength pixel selection matrix of potato seed potatoes is obtained by the above method and the eigenvalue vector , and the wavelength pixel selection matrix of normal potato seed potatoes as the reference standard and the eigenvalue vector Combining the F norm of the two matrices and the cosine values of the two eigenvalue vectors, the difference between the two is calculated by spectral angle matching as the potato seed potato disease index:
[0079] ;
[0080] In the formula, represents the potato seed potato disease index, represents the wavelength pixel selection matrix for potato seed potatoes, Represents the wavelength pixel selection matrix of normal potato seed potatoes, and They represent the number of pixel sequences in the wavelength pixel selection matrix of seed potato and normal potato, respectively. represents the F-norm of the matrix, and represent the eigenvalue vectors of seed potato and normal potato respectively, Indicates the modulus of the calculated vector, Indicates the number of bands.
[0081] It should be noted that when there is no disease in the seed potato, the hyperspectral data of its non-germination area is basically consistent with that of normal potato, and the wavelength pixel selection matrix obtained by both has fewer pixels screened out, so the F norm of the two wavelength pixel selection matrices is small, and the two eigenvalue vectors are not much different, so the final potato seed potato disease index is On the contrary, if there is disease in the seed potatoes, the seed potato tissue at the diseased location has deteriorated, so there is a certain difference between the hyperspectral data and the hyperspectral data of normal potatoes, and there are more abnormal pixels in the seed potatoes. At this time, both the wavelength pixel selection matrix and the eigenvalue vector are different, and the final potato seed potato disease index is Larger value.
[0082] Step S003, identifying the diseases of the seed potatoes in combination with the seed potato disease index.
[0083] The potato seed potato disease index was obtained by the method described in this example. , The value reflects the disease situation of potato seed potatoes. The larger the value, the more serious the disease of the seed potato.
[0084] Set the first threshold and the second threshold ;when When , the potato seed potatoes are healthy seed potatoes; when When , the potato seed potatoes are qualified, but there is a possibility of disease and further screening is required; when When the potato seed potatoes are diseased, the disease is serious and they need to be screened out from the seed potatoes to reduce the economic losses caused by the seed potato disease.
[0085] Among them, the first threshold and the second threshold In this embodiment, they are set to 0.1 and 0.2 respectively, and the first threshold is smaller than the second threshold, which can be set by the implementer.
[0086] At this point, the disease identification of potato seed potatoes can be completed.
[0087] In summary, the embodiment of the present invention proposes a method for identifying potato seed potato diseases based on high-resolution spectral features, collects hyperspectral data of potato seed potatoes, eliminates the interference of germination area pixels and redundant pixels on disease area pixels in turn, and constructs a potato seed potato disease index based on the characteristics of normal potato seed potatoes to complete potato seed potato disease identification.
[0088] The embodiment of the present invention divides seed potatoes into a germination area and a non-germination area, eliminates the influence of the germination area through semantic segmentation, constructs a non-germination area matrix, and excludes the identification of potato seed potato diseases in the germination area, so that the result is more accurate; the embodiment of the present invention calculates the wavelength variation evaluation coefficient for each pixel in the non-germination area matrix, constructs the wavelength variation matrix of the non-germination area, so that the highlight coefficient of each pixel in the non-germination area can be calculated, and a highlight pixel selection sequence is obtained. By screening out the highlight pixels, the pixels in the diseased area are screened out from the pixels in the normal area, and redundant pixels are eliminated, thereby reducing the time complexity of the algorithm;
[0089] The wavelength pixel selection matrix is obtained by combining the wavelength variation matrix of the non-germination area and the matrix of the non-germination area, and the corresponding eigenvalue vector is obtained, which can reflect the characteristics of normal seed potato. The difference between the wavelength pixel selection matrix and the eigenvalue vector of the normal seed potato is calculated to obtain the seed potato disease index, and the disease degree of the seed potato is indirectly reflected by the difference between the normal seed potato and the normal seed potato. The embodiment of the present invention eliminates the interference of the germinated area pixels and redundant pixels, and optimizes the matrix by technical means to obtain the corresponding eigenvalue vector, and finally constructs the seed potato disease index, avoids the calculation consumption caused by calculating a large number of redundant pixel sequences, improves the detection efficiency, and reduces the interference of pixel sequence noise by extracting features of the diseased area, thereby improving the detection accuracy.
[0090] It should be noted that the sequence of the above embodiments of the present invention is only for description and does not represent the advantages and disadvantages of the embodiments. The above is a description of a specific embodiment of this specification. In addition, the processes depicted in the accompanying drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.
[0091] The various embodiments in this specification are described in a progressive manner, and the same or similar parts between the various embodiments can be referenced to each other, and each embodiment focuses on the differences from other embodiments.
[0092] The embodiments described above are only used to illustrate the technical solutions of the present application, rather than to limit them. Modifications to the technical solutions recorded in the aforementioned embodiments, or equivalent replacement of some of the technical features therein, do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present application, and should all be included in the protection scope of the present application.
Claims
1. A method for identifying potato seed potato diseases based on high-resolution spectral features, characterized in that: The method comprises the following steps: Obtain hyperspectral data of potato seed potatoes; The hyperspectral data is synthesized into true color to obtain a pseudo color image; the pseudo color image is segmented into germination area and non-germination area by using semantic segmentation neural network; the non-germination area matrix is obtained based on the reflection intensity value of each pixel in the non-germination area in each band; the wavelength variation evaluation coefficient of each pixel is obtained according to the reflection intensity value of each pixel in the non-germination area matrix; the wavelength variation matrix of the non-germination area is obtained according to the wavelength variation evaluation coefficient of each pixel in the non-germination area matrix; the highlight coefficient of each row vector of the wavelength variation matrix of the non-germination area is calculated, and the highlight coefficients of all row vectors are arranged from large to small to construct a highlight wavelength selection sequence; The method for obtaining the highlight coefficient of each row vector is as follows: for each row vector of the wavelength variation matrix of the non-germination area, the mean value of the wavelength variation evaluation coefficients of all band pixels in the row vector is used as the highlight coefficient of the row vector; Adjust the non-germination area matrix and the non-germination area wavelength variation matrix according to the highlighted wavelength selection sequence; update the non-germination area wavelength variation matrix according to the wavelength variation evaluation coefficient of each pixel in the non-germination area wavelength variation matrix; obtain the wavelength pixel selection matrix according to the Hadamard product of the non-germination area matrix and the non-germination area wavelength variation matrix; obtain the eigenvalue vector of potato seed potatoes according to the wavelength pixel selection matrix; obtain the wavelength pixel selection matrix and eigenvalue vector of normal potato seed potatoes; A potato seed potato disease index is obtained according to the wavelength pixel selection matrix and eigenvalue vector of potato seed potatoes and normal potato seed potatoes; and the disease condition of potato seed potatoes is identified according to the potato seed potato disease index.
2. The method for identifying potato seed potato diseases based on high-resolution spectral features according to claim 1, characterized in that: The non-germination area matrix is obtained based on the reflection intensity value of each pixel in the non-germination area in each band, including: For each pixel in the non-germination area, the reflection intensity value of the pixel in each band is taken as the pixel sequence of the pixel; The pixel sequence of each pixel in the non-germination area is used as a row vector to form a non-germination area matrix.
3. The method for identifying potato seed potato diseases based on high-resolution spectral features according to claim 1, characterized in that: The wavelength variation evaluation coefficient of each pixel is obtained according to the reflection intensity value of each pixel in the non-germination area matrix, including: For each pixel in the matrix of the non-germination area, calculate the mean and standard deviation of the reflection intensity of all other pixels in the row vector where the pixel is located; The absolute value of the difference between the reflection intensity value of the pixel and the mean value of the reflection intensity is calculated, and the ratio of the absolute value of the difference to the standard deviation of the reflection intensity is used as the wavelength variation assessment coefficient of the pixel.
4. The method for identifying potato seed potato diseases based on high-resolution spectral features according to claim 1, characterized in that: The method of obtaining the wavelength variation matrix of the non-germination area according to the wavelength variation evaluation coefficient of each pixel of the non-germination area matrix includes: The reflection intensity value of each pixel in the non-germination area matrix was replaced with the wavelength variation evaluation coefficient of each pixel, and the obtained matrix was used as the wavelength variation matrix of the non-germination area.
5. The method for identifying potato seed potato diseases based on high-resolution spectral features according to claim 1, characterized in that: The method of adjusting the non-germination area matrix and the non-germination area wavelength variation matrix according to the highlighted wavelength selection sequence includes: The first m% pixels in the highlight wavelength selection sequence are used as the highlight pixels of the non-germination area, where m is the preset selection coefficient; The highlighted pixels in the non-germination area are screened out from the non-germination area matrix and the non-germination area wavelength variation matrix as the adjusted non-germination area matrix and the non-germination area wavelength variation matrix.
6. The method for identifying potato seed potato diseases based on high-resolution spectral features according to claim 1, characterized in that: The updating of the wavelength variation matrix of the non-germination area according to the wavelength variation evaluation coefficient of each pixel of the wavelength variation matrix of the non-germination area includes: For each pixel in the wavelength variation matrix of the non-germination area, when the wavelength variation assessment coefficient of the pixel is less than the screening threshold, the wavelength variation assessment coefficient of the pixel is set to zero; when the wavelength variation assessment coefficient of the pixel is greater than or equal to the screening threshold, the wavelength variation assessment coefficient of the pixel remains unchanged.
7. The method for identifying potato seed potato diseases based on high-resolution spectral features according to claim 1, characterized in that: The step of obtaining the eigenvalue vector of seed potato according to the wavelength pixel selection matrix includes: Singular value decomposition is used to calculate the eigenvalues of the wavelength pixel selection matrix, and all eigenvalues are combined into the eigenvalue vector of potato seed potatoes.
8. The method for identifying potato seed potato diseases based on high-resolution spectral features according to claim 1, characterized in that: The potato seed potato disease index is obtained according to the wavelength pixel selection matrix and eigenvalue vector of the potato seed potato and the normal potato seed potato, and the expression is: ; In the formula, represents the potato seed potato disease index, represents the wavelength pixel selection matrix for potato seed potatoes, Represents the wavelength pixel selection matrix of normal potato seed potatoes, and Respectively represent the number of pixel sequences in the wavelength pixel selection matrix of seed potato and normal potato, represents the F-norm of the matrix, and represent the eigenvalue vectors of seed potato and normal potato respectively, Indicates the modulus of the calculated vector, Indicates the number of bands.
9. The method for identifying potato seed potato diseases based on high-resolution spectral features according to claim 1, characterized in that: The method of identifying the disease condition of seed potatoes according to the seed potato disease index includes: When the potato seed potato disease index is less than or equal to the first threshold, the potato seed potatoes are identified as healthy seed potatoes; When the potato seed potato disease index is greater than the first threshold value and less than or equal to the second threshold value, the potato seed potatoes are identified as qualified seed potatoes; When the potato seed potato disease index is greater than the second threshold, the potato seed potatoes are identified as diseased seed potatoes, wherein the first threshold is less than the second threshold.
