Identification method of asymptomatic infection of strawberry anthracnose based on hyperspectral features

Abstract

The present invention discloses a method for identifying asymptomatic anthracnose infection in strawberries based on hyperspectral features. The method specifically includes collecting raw spectral feature data and preprocessing the spectral data, screening characteristic wavelengths for detecting asymptomatic anthracnose infection in strawberries, and constructing asymptomatic anthracnose identification models for strawberries using random forests and BP neural networks based on the characteristic wavelengths. The constructed models can non-destructively, quickly, and accurately identify anthracnose-infected strawberry plants during the asymptomatic period. The model constructed based on random forests has a higher recognition rate, while the model constructed using the BP neural network recognition model has higher recognition efficiency and faster speed. In production, an appropriate strawberry disease identification model can be selected according to different needs. The present invention achieves a breakthrough in strawberry disease identification, saves time and the cost of traditional biochemical analysis, improves economic benefits, achieves cost reduction, and conforms to the green and sustainable development of agriculture.

Description

Technical Field

[0001] The present invention specifically relates to the field of antenna technology, and in particular to a method for identifying asymptomatic infection of strawberry anthracnose based on hyperspectral features. Background Art

[0002] Strawberry anthracnose occurs in strawberry-growing areas worldwide, causing significant damage to both nursery and production fields, with the two being interconnected. Anthracnose in strawberry nursery fields not only causes seedling losses, but also leads to the emergence of a certain number of asymptomatic diseased seedlings due to the latent infection of anthracnose. Once these diseased seedlings are planted in production fields, the disease accelerates and spreads, leading to seedling mortality. This has been a major cause of early seedling mortality in strawberry production fields since the introduction of the 'Hongyan' strawberry, especially over the past decade. Currently, identification of strawberry anthracnose primarily relies on manual identification and laboratory PCR testing. While experienced strawberry farmers can identify symptomatic anthracnose seedlings, not all possess sufficient experience, and empirical evidence alone is insufficient to identify asymptomatic carriers. PCR technology can detect the presence of anthracnose in strawberry seedlings, but it has several limitations. First, testing requires sampling of leaves, petioles, or root collars, which damages the plant to a certain extent. Second, due to cost constraints, only sampling can be used, making it impossible to determine the presence of individual plants. Third, testing relies on laboratories, making it difficult for strawberry producers to implement. Fourth, testing is time-consuming and not always effective. Currently, there is no rapid, non-destructive method for identifying asymptomatic anthracnose infections in strawberries. Therefore, rapid diagnosis and detection of asymptomatic anthracnose infections is crucial for strawberry disease prevention and control. Summary of the Invention

[0003] The purpose of the present invention is to overcome the defects of the above-mentioned prior art and propose a method for identifying asymptomatic infection of strawberry anthracnose based on hyperspectral features, thereby realizing asymptomatic infection detection of strawberry anthracnose.

[0004] To achieve the purpose of the present invention, the present invention proposes a method for identifying asymptomatic infection of strawberry anthracnose based on hyperspectral features, which is characterized by comprising the following steps:

[0005] 1) Preparing activated anthrax bacteria and healthy strawberry plants for inoculation, placing the infected strawberry plants and healthy strawberry plants in a hyperspectral acquisition device to collect raw hyperspectral data;

[0006] 2) Perform black and white image correction on the acquired original hyperspectral data;

[0007] 3) Denoising the data after black and white image correction;

[0008] 4) For the denoised image, extract the root neck area of ​​the healthy plant as the healthy region of interest, and obtain the average spectrum of all healthy regions of interest; extract the diseased area of ​​the diseased plant as the diseased region of interest, and obtain the average spectrum of all diseased regions of interest; extract the asymptomatic infection area outside the diseased area, which is infected but has no phenotypic characteristics, and obtain the average spectrum of all asymptomatic infections; the extraction method of the asymptomatic infection area is as follows:

[0009] 4.1. Select several pixels within a 2mm radius of the diseased area as candidate areas for asymptomatic infection, avoiding diseased pixels that are black or brown in color.

[0010] 4.2. Extract the average spectrum of pixels in the candidate asymptomatic infection area and observe the spectral characteristic curve of the average spectrum;

[0011] 4.3. Eliminate healthy pixels from the candidate asymptomatic infection area. The criteria for eliminating healthy pixels are as follows: After selecting the candidate asymptomatic infection area, observe the spectral characteristics. If a pixel has obvious troughs in the intervals of 530-550nm, 580-620nm, and 656-700nm, it is determined to be a healthy pixel and eliminated. The remaining pixels constitute the asymptomatic infection area.

[0012] 5) Perform SG smoothing preprocessing on the average spectrum of all regions of interest to further eliminate the influence of sample surface unevenness and noise on the data;

[0013] 6) For the data set obtained after SG smoothing preprocessing, the SPA continuous projection algorithm is used to select the characteristic wavelength;

[0014] 7) The characteristic wavelength data of the healthy region of interest, the diseased region of interest, and the asymptomatic infection region are divided into a training set T1 and a test set T2 respectively;

[0015] 8) Inputting the spectral data of training set T1 into BP neural network and random forest to establish a strawberry disease severity discrimination model;

[0016] 9) Inputting the spectral data of the test set T2 into the strawberry disease severity discrimination model established in step 8) to verify the accuracy of the strawberry disease severity discrimination model;

[0017] 10) Obtain hyperspectral data of the strawberry plant to be tested and perform black and white image correction, and then use the strawberry disease severity discrimination model verified in step 9) to identify the strawberry plant. If there is an asymptomatic infection area in the identification result, the strawberry plant to be tested has been infected with anthracnose.

[0018] Compared with the prior art, the present invention has the following advantages:

[0019] 1. After the discrimination model is established, the present invention only needs to input the characteristic wavelength data extracted by the SPA algorithm to obtain the test results, which can achieve non-destructive, rapid and accurate identification of strawberry anthracnose.

[0020] 2. The present invention combines the characteristics of spectral curves to achieve accurate extraction of asymptomatic infection areas, and uses them together with healthy areas and diseased areas as three types of samples for machine learning, thereby realizing rapid diagnosis and detection of asymptomatic infection of strawberry anthracnose.

[0021] 3. Using SPA continuous projection algorithm, the calculation is simpler and the effect of screening characteristic wavelength combinations is better.

[0022] 4. From data collection to modeling and analysis, the time for disease identification is greatly shortened and the efficiency of disease identification is improved compared with traditional biochemical detection and analysis. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] Figure 1 The present invention is a flow chart of the method for identifying asymptomatic infection of strawberry anthracnose.

[0024] Figure 2 The average spectrum of healthy, asymptomatic and symptomatic strawberries.

[0025] Figure 3 Average spectra of the region of interest extracted for all samples.

[0026] Figure 4 This is the spectral curve after SG preprocessing.

[0027] Figure 5 The results of extracting characteristic bands using the SPA continuous projection algorithm, (a) shows the six characteristic wavelengths extracted by the SPA continuous projection algorithm, and (b) shows the distribution of the six extracted characteristic wavelengths.

[0028] Figure 6 is the confusion matrix of the test set discrimination result, (a) is the confusion matrix of the random forest discrimination result, and (b) is the confusion matrix of the BP neural network discrimination result. DETAILED DESCRIPTION

[0029] The present invention will be further explained below in detail with reference to the accompanying drawings so that those skilled in the art can have a deeper understanding of the present invention and be able to implement it. However, the following reference examples are only used to explain the present invention and are not intended to limit the present invention.

[0030] like Figure 1 FIG. 1 is a flow chart of a method for identifying asymptomatic infection of strawberry anthracnose based on hyperspectral features according to an embodiment of the present invention. The method of the present invention specifically comprises the following steps:

[0031] 1) Prepare activated anthrax bacteria and healthy strawberry plants for pathogen inoculation, place the infected strawberry plants and healthy strawberry plants in a hyperspectral acquisition device, and collect original hyperspectral data. In this step, the infected plants and healthy plants are strawberry plants of the same period and variety to eliminate the interference of time and variety factors. The hyperspectral data acquisition system consists of a spectral camera, a light source, a double-branched light guide, a dark box, a computer, and an electrically controlled shift platform. The spectrum wavelength range collected is 373-1033nm. In order to adjust the equipment parameters for accuracy in taking pictures, a 35mm imaging lens is selected, and the adjustment parameter values ​​determine that the object distance is 27cm, the exposure time is 2.0ms, and the light source intensity is 100lx.

[0032] 2) Perform black and white image correction on the acquired original hyperspectral data. The black and white image correction is completed based on the following formula:

[0033]

[0034] Where R(i) is the relative reflectivity of pixel i in the corrected image, I r (i) is the reflectance of pixel i in the original image of the test spectrum, I w (i) is the spectral reflectance of pixel i in the image after whiteboard correction, I d (i) is the spectral reflectance of pixel i in the image after blackboard correction.

[0035] 3) The original data has a lot of noise at the beginning and end, so the data after the black and white image correction needs to be denoised. The data denoising method in this embodiment uses band cutting to remove the bands with large noise at the beginning and end. The spectral range after band cutting is 447-965nm, which achieves preliminary noise elimination.

[0036] 4) For the denoised image, extract the root neck area of ​​the healthy plant as the healthy region of interest, and obtain the average spectrum of all healthy regions of interest; extract the diseased area of ​​the diseased plant as the diseased region of interest, and obtain the average spectrum of all diseased regions of interest; extract the asymptomatic infection area near the diseased area where the disease infection exists but the phenotype is uncharacteristic, and obtain the average spectrum of all asymptomatic infections. Use the ROI extraction function in the ENVI software to extract the healthy region of interest, the diseased region of interest, and the asymptomatic infection region. Extract 1200 healthy, asymptomatic, and infected samples, totaling 3600 samples. The average spectral curve of the region of interest extracted from all samples is shown as follows. Figure 3 shown.

[0037] The method for extracting asymptomatic infection areas is as follows:

[0038] 4.1. Select several pixels within 2 mm of the diseased area as candidate areas for asymptomatic infection, avoiding selecting diseased pixels that are black or brown in color.

[0039] 4.2. Extract the average spectrum of pixels in the candidate asymptomatic infection area and observe the spectral characteristic curve of the average spectrum; Figure 2 Shown is the average spectral curve of all healthy, asymptomatic and symptomatic infected samples obtained: the solid line represents healthy samples, the dotted line represents asymptomatic infected samples, and the dot-dash line represents symptomatic infected samples. It can be seen from the whole figure that the trend of the healthy sample curve is significantly different from that of the infected sample curve. There are obvious troughs between 530-550nm, 580-620nm and 656-700nm for healthy samples, while there are no obvious troughs at these locations for the infected bacteria samples. Although the trends of the curves of asymptomatic and symptomatic infected samples are similar, it can be observed that as the disease deepens, the anthrax infection changes from asymptomatic infection to symptomatic infection, and the spectral reflectance is gradually decreasing (the spectral changes during the development of the disease are based on changes in plant pigments, water and nutrient content, leading to the above-mentioned infection symptoms).

[0040] 4.3. Eliminate healthy pixels from the candidate asymptomatic infection area. The conditions for eliminating healthy pixels are as follows: After selecting the candidate asymptomatic infection area, observe the spectral characteristics. If the pixel has obvious troughs in the intervals of 530-550nm, 580-620nm, and 656-700nm, it is determined to be a healthy pixel and eliminated. The remaining pixels constitute the asymptomatic infection area.

[0041] 5) Perform Savitzky-Golay smoothing filtering on the average spectrum of all regions of interest to achieve spectral preprocessing, further eliminating the impact of sample surface unevenness and noise on the data. The SG smoothing window size is selected as 5, and the spectral curve after preprocessing is as follows Figure 4 shown.

[0042] 6) For the data set obtained after SG smoothing preprocessing, the SPA continuous projection algorithm is used to select the characteristic wavelength. The SPA continuous projection algorithm first selects a spectral column vector in the spectral data as the starting vector; then calculates the projections of other column vectors on the starting vector; then selects the minimum projection as the starting vector of the next projection; until the number of selected variables reaches the input iteration number; finally, all the extracted wavelength combinations are subjected to multivariate linear regression, and the band combination with the highest prediction accuracy is selected from the local minimum root mean square error as the optimal characteristic band combination. The SPA continuous projection algorithm is within the scope of existing technology and will not be described in detail in this embodiment. The characteristic wavelength combinations finally obtained are 654nm, 673nm, 593nm, 736nm, 965nm, and 447nm. The SPA algorithm extraction process and results are shown in the figure. Figure 5 shown.

[0043] 7) The characteristic wavelength data of the healthy region of interest, the diseased region of interest, and the asymptomatic infection region are divided into a training set T1 and a test set T2 in a ratio of 3:1.

[0044] 8) The spectral data from training set T1 were input into the BP neural network and random forest classification systems, respectively. The BP neural network had a maximum iteration count of 1000, a learning rate of 0.01, and a network structure with 6 nodes in the input layer, 6 nodes in the hidden layer, and 3 nodes in the output layer. The number of decision trees in the random forest was set to 500. Thus, a model for distinguishing the severity of strawberry disease (BP neural network and random forest classifier) ​​was established.

[0045] 9) Inputting the spectral data of the test set T2 into the two strawberry disease severity discrimination models established in step 8) to verify the accuracy of the strawberry disease severity discrimination models.

[0046] Table 1 shows the confusion matrix of the discrimination results of the random forest and BP neural network test sets.

[0047]

[0048] The model established in this experiment has a recognition rate of 93.7% and 93.2% for the three different types of samples in different models, respectively, which shows a good discrimination effect. This shows that the model established by the method of the present invention can effectively identify the three types of samples: healthy, asymptomatic infection, and symptomatic infection. Figure 6 Shown is the confusion matrix of the discrimination results of the test set.

[0049] 10) Obtain hyperspectral data of the strawberry plant to be tested and perform black and white image correction, and then use the strawberry disease severity discrimination model verified in step 9) to identify the strawberry plant. If there is an asymptomatic infection area in the identification result, the strawberry plant to be tested has been infected with anthracnose.

[0050] The specific implementation scheme described above further illustrates in detail the purpose, technical solutions and beneficial effects of the present invention. It should be understood that the above is only a specific implementation scheme of the present invention and is not intended to limit the scope of the present invention. Any equivalent changes and modifications made by any technician in this field without departing from the concept and principle of the present invention should fall within the scope of protection of the present invention.

Claims

1. A method for identifying asymptomatic infection of strawberry anthracnose based on hyperspectral features, characterized in that: The following steps are involved: 1) Preparing activated anthrax bacteria and healthy strawberry plants for inoculation, placing the infected strawberry plants and healthy strawberry plants in a hyperspectral acquisition device to collect raw hyperspectral data; 2) Perform black and white image correction on the acquired original hyperspectral data; 3) Denoising the data after black and white image correction; 4) For the denoised image, extract the root neck area of ​​the healthy plant as the healthy region of interest, and obtain the average spectrum of all healthy regions of interest; extract the diseased area of ​​the diseased plant as the diseased region of interest, and obtain the average spectrum of all diseased regions of interest; extract the asymptomatic infection area outside the diseased area, which is infected but has no phenotypic characteristics, and obtain the average spectrum of all asymptomatic infections; the extraction method of the asymptomatic infection area is as follows: 4.

1. Select several pixels within a 2mm radius of the diseased area as candidate areas for asymptomatic infection, avoiding diseased pixels that are black or brown in color. 4.

2. Extract the average spectrum of pixels in the candidate asymptomatic infection area and observe the spectral characteristic curve of the average spectrum; 4.

3. Eliminate healthy pixels from the candidate asymptomatic infection area. The criteria for eliminating healthy pixels are as follows: After selecting the candidate asymptomatic infection area, observe the spectral characteristics. If a pixel has obvious troughs in the intervals of 530-550nm, 580-620nm, and 656-700nm, it is determined to be a healthy pixel and eliminated. The remaining pixels constitute the asymptomatic infection area. 5) Perform SG smoothing preprocessing on the average spectrum of all regions of interest to further eliminate the influence of sample surface unevenness and noise on the data; 6) For the data set obtained after SG smoothing preprocessing, the SPA continuous projection algorithm is used to select the characteristic wavelength; 7) The characteristic wavelength data of the healthy region of interest, the diseased region of interest, and the asymptomatic infection region are divided into a training set T1 and a test set T2 respectively; 8) Inputting the spectral data of the training set T1 into the classifier to establish a strawberry disease severity discrimination model; 9) Inputting the spectral data of the test set T2 into the strawberry disease severity discrimination model established in step 8) to verify the accuracy of the strawberry disease severity discrimination model; 10) Obtain hyperspectral data of the strawberry plant to be tested and perform black and white image correction, and then use the strawberry disease severity discrimination model verified in step 9) to identify the strawberry plant. If there is an asymptomatic infection area in the identification result, the strawberry plant to be tested has been infected with anthracnose.

2. A method for identifying asymptomatic infection of strawberry anthracnose based on hyperspectral features according to claim 1, characterized in that: In step 1), the diseased plants and the healthy plants are strawberry plants of the same period and the same variety to eliminate the interference of time and variety factors.

3. A method for identifying asymptomatic infection of strawberry anthracnose based on hyperspectral features according to claim 1, characterized in that: In step 1), the hyperspectral data acquisition system consists of a spectral camera, a light source, a double-branched light guide, a darkroom, a computer, and an electrically controlled shift platform. The collected spectrum wavelength range is 373-1033 nm. A 35 mm imaging lens is selected, and the parameter values ​​are adjusted to determine the object distance to be 27 cm, the exposure time to be 2.0 ms, and the light source intensity to be 100 lx.

4. A method for identifying asymptomatic infection of strawberry anthracnose based on hyperspectral features according to claim 1, characterized in that: In step 2), the black and white image correction is performed based on the following formula: Where R(i) is the relative reflectivity of pixel i in the corrected image, I r (i) is the reflectance of pixel i in the original image of the test spectrum, I w (i) is the spectral reflectance of pixel i in the image after whiteboard correction, I d (i) is the spectral reflectance of pixel i in the image after blackboard correction.

5. The method for identifying asymptomatic infection of strawberry anthracnose based on hyperspectral features according to claim 1, wherein: In step 3), the data denoising method is to remove the bands with large noise at the beginning and end by band cutting. The spectral range after band cutting is 447-965nm, which achieves preliminary noise elimination.

6. The method for identifying asymptomatic infection of strawberry anthracnose based on hyperspectral features according to claim 1, wherein: In step 4), the ROI extraction function in ENVI software was used to extract healthy regions of interest, diseased regions of interest, and asymptomatic infection regions.

7. The method for identifying asymptomatic infection of strawberry anthracnose based on hyperspectral features according to claim 1, wherein: In step 6), the SPA continuous projection algorithm is used to select the characteristic wavelengths. First, a spectral column vector is selected in the spectral data as the starting vector; then the projections of other column vectors on the starting vector are calculated; the minimum projection is selected as the starting vector of the next projection; until the number of selected variables reaches the input iteration number; finally, all the extracted wavelength combinations are subjected to multivariate linear regression, and the band combination with the highest prediction accuracy is selected from the local minimum root mean square error as the optimal characteristic band combination.

8. The method for identifying asymptomatic infection of strawberry anthracnose based on hyperspectral features according to claim 7, wherein: The final characteristic wavelength combination is 654nm, 673nm, 593nm, 736nm, 965nm, and 447nm.

9. The method for identifying asymptomatic infection of strawberry anthracnose based on hyperspectral features according to claim 1, wherein: In step 8), the classifier selects BP neural network or random forest classifier.

10. The method for identifying asymptomatic infection of strawberry anthracnose based on hyperspectral features according to claim 9, characterized in that: In step 8), the maximum number of iterations of the BP neural network is 1000, the learning rate is 0.01, and the network structure used is: the input layer has 6 nodes, the hidden layer has 6 nodes, and the output layer has 3 nodes; the number of decision trees in the random forest is set to 500.

Images