Method for rapidly identifying acid resistance of rape variety based on RGB image

Abstract

The invention relates to the field of agricultural artificial intelligence and crop breeding and breeding, and discloses a method for rapidly identifying the acid resistance of a rape variety based on an RGB image. The method comprises the following steps: respectively planting to-be-screened rape varieties in acid soil and neutral soil, regularly collecting RGB images of a top view and a side view in the whole growth period of rape, extracting a plant area by using a deep learning segmentation algorithm, and obtaining texture features and morphological features as feature values; the change rate of characteristic values under acidic and neutral soil is calculated, an acid stress discrimination model is established in combination with a machine learning algorithm, key characteristic values are screened, and the weight of the characteristic values is determined. And calculating an acid resistance evaluation value and a time node weight of each time point to obtain a comprehensive acid resistance evaluation value in the whole growth period, thereby realizing accurate and rapid screening of the acid resistance of the rapes. The method provided by the invention is efficient and accurate in operation, avoids tedious and subjective errors of manual measurement, is helpful for accelerating the breeding process, and is of great significance to acid-resistant genetic improvement and breeding research of oilseed rapes.

Description

Technical Field

[0001] The present invention relates to the fields of agricultural artificial intelligence and crop breeding and selection, and in particular to a method for quickly identifying the acid resistance of rapeseed varieties based on RGB images. Background Art

[0002] Rapeseed is the world's third largest oilseed crop and an important raw material for vegetable oil. It has great value in a variety of food and non-food industries. Rapeseed is an important high-quality oil crop in my country, especially in the middle and lower reaches of the Yangtze River and the Sichuan Basin, where rapeseed is a mainstay cash crop for local farmers. my country's acidic soils are mainly distributed in the vast tropical and subtropical areas south of the Yangtze River and in Yunnan, Guizhou and Sichuan, covering an area of ​​about 200 million hectares, which overlaps with the main rapeseed planting areas. The loss of nutrients such as potassium, calcium, and magnesium and heavy metal poisoning caused by soil acidification seriously affect the activity of soil micro-soil enzymes and nutrient effectiveness, which will eventually lead to a reduction in rapeseed production and has become a limiting factor for agricultural production in some regions.

[0003] Traditional soil improvement methods in agricultural production, such as applying quicklime and other conditioners to acidic soils, can neutralize the acidity and alkalinity of the soil to a certain extent. However, due to the high cost of quicklime itself and the fact that it easily causes environmental pollution and a certain degree of soil compaction, it has great limitations in improving acidic environments. During the promotion of modern crop breeding technology, it was found that there are significant genotypic differences in the acid resistance of different varieties of rapeseed. Therefore, high-throughput phenotyping and genetic breeding technology can be used to screen and cultivate acid-resistant and high-yield rapeseed varieties, breaking through the limitation of low yield in acidic soils.

[0004] Currently, there are few methods for screening acid-resistant rapeseed, and existing screening methods are mostly focused on the tolerance screening of other crops or the screening of rapeseed under other abiotic stress conditions. Existing technologies usually rely on manual measurement of plant morphological indicators, such as root length, leaf length, plant height, number of siliques, etc. These methods are not only relatively cumbersome and time-consuming in the measurement process, but also may be destructive to the plants. In addition, although some methods combine multi-stage screening processes, the screening steps are complicated and require high human resources. In the screening of rapeseed under abiotic stress, common problems also include that the evaluation indicators are too single and cannot fully reflect the comprehensive effects of stress conditions on plants. Especially under acidic stress conditions, due to the lack of obvious specific traits of rapeseed plants, the methods for screening tolerant varieties under other stress conditions cannot be directly applied to acid resistance screening, which limits the selection of rapeseed varieties with strong adaptability in acidic soil environments.

[0005] Therefore, in order to alleviate the limitations of acidic soil on agricultural production and accelerate the breeding research and promotion of acid-resistant rapeseed varieties, an efficient and accurate method for screening acid-resistant rapeseed varieties is needed. Summary of the invention

[0006] The technical problems to be solved by the present invention include but are not limited to the following: the existing rapeseed stress tolerance screening method relies on manual measurement and has a single evaluation index; the prior art lacks an efficient and accurate rapeseed variety screening method directly applicable to acidic stress conditions.

[0007] In order to solve the above problems, the present invention provides the following technical solutions.

[0008] A method for quickly identifying the acid resistance of rapeseed varieties based on RGB images comprises the following steps: Step S1, dividing the rapeseed varieties to be screened into two treatment groups for potted planting, one group is placed in acidic soil, and the other group is placed in neutral soil, and other physical and chemical properties of the two groups of soils are kept consistent; Step S2, starting from the rapeseed bud stage, regularly using an RGB image acquisition device to simultaneously acquire top view and side view images until the end of the entire growth period; Step S3, segmenting the rapeseed plant regions in the top view and the side view by a deep learning algorithm, and extracting texture features and morphological features from the segmented images as feature values; Step S4, at each image acquisition time point, for all tested varieties, the characteristic value change rates between adjacent time points under acidic and neutral soil environments are calculated respectively, an acid stress discrimination model is established using a machine learning algorithm, key characteristic values ​​are screened and their weights are determined, and the acid resistance evaluation value of each variety at that time point is calculated based on the key characteristic value change rate and weight under the acidic soil environment; Step S5, based on the acid resistance evaluation value at each time point in the whole growth period obtained in step S4, analyze the importance of the acid resistance evaluation value at different time nodes to the overall evaluation, determine the time node weight, and combine the acid resistance evaluation value at each time point and its time node weight to calculate the comprehensive acid resistance evaluation value for the whole growth period, which is used to judge the acid resistance of rapeseed varieties. The larger the comprehensive evaluation value, the stronger the acid resistance.

[0009] Preferably, the image segmentation in step S3 is implemented in the following manner: first, the original image is labeled to obtain a binary image label, the training set and the test set are divided according to a certain ratio, and then the Segformer segmentation network is used to train the data set, and the optimal weight of the network is derived to segment all rapeseed images.

[0010] Preferably, the texture features are extracted through grayscale histogram and grayscale co-occurrence matrix, and the obtained parameters include but are not limited to correlation, gradient advantage, energy, gradient mean, gradient entropy, differential moment, gradient standard deviation and third-order moment.

[0011] Preferably, the morphological features are calculated by processing a binary image of rapeseed, and the obtained parameters include but are not limited to width, height, projected area, convex hull area, perimeter-to-area ratio, compactness and fractal dimension.

[0012] Preferably, in step S4, the calculation formula of the characteristic value change rate is: in, It represents the rate of change of the eigenvalue with eigenvalue index f between adjacent time points t+1 and t; represents the feature value with index f extracted from the image collected at time point t+1, It means that the index extracted from the image collected at time point t is the f eigenvalue, where F is the sum of the eigenvalue index sets, including all the eigenvalues ​​used for calculation.

[0013] Preferably, in step S4, a random forest algorithm is used to establish an acid stress discrimination model, wherein the input of the acid stress discrimination model is the characteristic value change rate of the variety under the acidic soil environment and the neutral soil environment, and the output is the environment type; the model calculates the influence of each feature on the classification result by a feature importance measurement method, and uses it as the characteristic value weight w f Used for calculation of acid resistance evaluation value.

[0014] Preferably, in step S4, the acid resistance evaluation value R at a single time point is t The calculation formula is: .

[0015] Preferably, in step S5, the time node weight w t The calculation method is to perform principal component analysis on the acid resistance evaluation values ​​at different time points to obtain the principal component eigenvalues ​​and eigenvectors, where the principal component eigenvalues ​​represent the variance that the principal component can explain; then calculate the proportion of each principal component eigenvalue to the sum of all principal component eigenvalues, that is, the explained variance ratio; finally, weight the components of all principal component eigenvectors and the corresponding explained variance ratios to obtain the weight w at each time point. t .

[0016] Preferably, in step S5, the comprehensive evaluation value of acid resistance during the whole growth period ARS total The calculation formula is: Among them, total indicates the total number of time points.

[0017] Compared with the prior art, the method for rapidly identifying the acid resistance of rapeseed varieties based on RGB images provided by the present invention has significant positive technical effects, which are specifically manifested in the following aspects.

[0018] (1) Currently, there are relatively few screening methods for acid-resistant rapeseed varieties, and different abiotic stresses have different impact mechanisms on rapeseed. Under acidic stress conditions, rapeseed plants usually lack significant specific traits, so other stress screening methods such as salt tolerance and drought tolerance cannot be directly applied to acid resistance screening. In view of the uniqueness of acidic stress, the present invention designs a new rapid identification method, which makes up for the shortcomings of the existing technology and provides new ideas and technical support for rapeseed acid resistance research and variety development.

[0019] (2) In the prior art, manual data measurement is often used to screen rapeseed varieties under stress environments. Usually, during the maturity period of rapeseed, parameters such as plant height, leaf length, and leaf area are measured, and then its performance is comprehensively evaluated. However, this method is not only time-consuming and labor-intensive in the data collection process, but also requires a large amount of human resources for large-scale variety screening, and is inefficient. Compared with traditional manual observation methods, the present invention avoids the human errors that may be caused by traditional manual observation. The use of a unified data collection and processing process makes the data extraction of all samples more efficient and accurate, significantly reduces labor costs, and greatly improves the reliability of screening work.

[0020] (3) The existing technology usually only considers a single indicator in the evaluation of rapeseed varieties' acid resistance, which cannot fully reflect the actual performance of the plants. The present invention introduces a multi-time point data collection method combined with principal component analysis technology to comprehensively evaluate the importance of different growth stages and different characteristics. For example, in the seedling stage, more attention is paid to the leaf crown area in the top view, while in the mature stage, the focus is on evaluating plant height and biomass. This multi-dimensional evaluation method breaks through the limitations of a single indicator, improves the scientificity and accuracy of rapeseed's acid resistance evaluation, can provide more comprehensive screening results, and provide an important reference for rapeseed variety improvement. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] Figure 1 This is a flow chart of the method for screening acid-resistant rapeseed varieties.

[0022] Figure 2 This is a graph of eigenvalue weights determined based on the random forest algorithm.

[0023] Figure 3 Top and side views of acid-tolerant rapeseed varieties for screening.

[0024] Figure 4 Top and side views of sensitive rapeseed varieties for screening. DETAILED DESCRIPTION

[0025] The present invention is further described below in conjunction with the accompanying drawings and implementation examples.

[0026] like Figure 1The method flow chart shown is a preferred embodiment of the present invention, a method for quickly identifying the acid resistance of rapeseed varieties based on RGB images, comprising the following steps: Step S1, the rapeseed varieties to be screened are divided into two treatment groups for potted planting, one group is placed in acidic soil and the other group is placed in neutral soil. The acidic soil is taken from the red soil of Shuangxiqiao Town, Xianning City, Hubei Province. After repeated sampling and determination, the soil pH is 5.0. The second group adds 0.25% calcium hydroxide powder to the acidic soil and mixes it thoroughly to adjust the soil pH to 6.5. The other physical and chemical properties of the two groups of soils remain consistent; Step S2, starting from the rapeseed bud stage, regularly use RGB image acquisition equipment to simultaneously collect top view and side view images until the end of the full growth period; collect once every 14 days, and collect images in a dark room with a stable light source, using a black cloth as the background during shooting. Take a vertical shot of the top view, and take a side view from the side at the same time, to ensure that the time points of the two sets of acquisitions are consistent; Step S3, extracting rapeseed plant regions in the top view and the side view by a deep learning algorithm, and extracting texture features and morphological features from the segmented images as feature values; Step S4, at each image acquisition time point, for all tested varieties, the characteristic value change rates between adjacent time points under acidic and neutral soil environments are calculated respectively, an acid stress discrimination model is established using a machine learning algorithm, key characteristic values ​​are screened and their weights are determined, and the acid resistance evaluation value of each variety at that time point is calculated based on the key characteristic value change rate and weight under the acidic soil environment; Step S5, based on the acid resistance evaluation value at each time point in the whole growth period obtained in step S4, analyze the importance of the acid resistance evaluation value at different time nodes to the overall evaluation, determine the time node weight, and combine the acid resistance evaluation value at each time point and its time node weight to calculate the comprehensive acid resistance evaluation value for the whole growth period to judge the acid resistance of the rapeseed variety, wherein the larger the comprehensive evaluation value, the stronger the acid resistance.

[0027] The image segmentation steps in step S3 are as follows.

[0028] (1) Dataset preparation: A portion of the top view and side view images obtained in step S2 was randomly selected and annotated using the Eiseg software. During the annotation process, the image segmentation mode was selected.

[0029] (2) Model training: Use the SegFormer network in the paddleseg framework to train the dataset produced in step (1). After the model is fitted, the optimal model weights of the segmentation network are derived.

[0030] (3) Model prediction: Use the optimal model weights of the segmentation network obtained in step (2) to segment all the rapeseed side views and top views.

[0031] In step S3, the texture features are extracted through grayscale histogram and grayscale co-occurrence matrix, and the obtained parameters include but are not limited to correlation, gradient advantage, energy, gradient mean, gradient entropy, differential moment, gradient standard deviation, and third-order moment.

[0032] In step S3, the morphological features are calculated by processing the rapeseed binary image, and the obtained parameters include but are not limited to width, height, projected area, convex hull area, perimeter-to-area ratio, compactness and fractal dimension.

[0033] In step S4, the calculation formula of the characteristic value change rate is: in, It represents the rate of change of the eigenvalue with eigenvalue index f between adjacent time points t+1 and t; represents the feature value with index f extracted from the image collected at time point t+1, It means that the index extracted from the image collected at time point t is the f eigenvalue, where F is the sum of the eigenvalue index sets, including all the eigenvalues ​​used for calculation.

[0034] In step S4, a random forest algorithm is used to establish an acid stress discrimination model, the input of which is the characteristic value change rate of the variety under acidic soil environment and neutral soil environment, and the output is the environment type; the model calculates the influence of each feature on the classification result by the Gini index decline method, and uses it as the characteristic value weight w f Used for calculation of acid resistance evaluation value, the result is as follows Figure 2 shown.

[0035] In step S4, the acid resistance evaluation value R at a single time point t The calculation formula is: .

[0036] Time node weight w t The calculation method is to perform principal component analysis on the acid resistance evaluation values ​​at different time points to obtain the principal component eigenvalues ​​and eigenvectors, where the principal component eigenvalues ​​represent the variance that the principal component can explain; then calculate the proportion of each principal component eigenvalue to the sum of all principal component eigenvalues, that is, the explained variance ratio; finally, weight the components of all principal component eigenvectors and the corresponding explained variance ratios to obtain the weight w at each time point. t .

[0037] In step S5, the comprehensive evaluation value of acid resistance during the whole growth period ARS total The calculation formula is: Among them, total indicates the total number of time points.

[0038] Finally, 10 acid-resistant varieties and 10 sensitive varieties were selected based on the comprehensive evaluation value of acid resistance during the entire growth period. Figure 3 Shown are the side and top views of one of the acid-tolerant varieties, whose growth is not significantly affected by the acidic environment. Figure 4 The side view and top view of one of the sensitive varieties screened out are shown, indicating that the growth of this variety in an acidic environment is significantly inhibited and performs worse than in a neutral environment.

[0039] The total number of varieties to be tested in this example is 65, and the average absolute yield under acidic environment is 6.1399g / plant, and the average yield reduction rate is 43.65%. The absolute yield of the acid-resistant varieties screened out under acidic conditions is 7.1122g / plant, which is 15.83% higher than that of the varieties to be tested, and the yield reduction rate is 36.09%, which is 7.56% lower than that of the varieties to be tested. The absolute yield of sensitive varieties is 5.1545 g / plant, which is 16.05% lower than that of the varieties to be tested, and the yield reduction rate is 49.47%, which is 5.82% higher than that of the varieties to be tested. The acid-resistant varieties screened out by the present invention show good adaptability under acidic environments, and the yield of the varieties to be tested under acidic environments is improved, and the yield reduction rate is reduced. These results show that the screening method proposed in the present invention can effectively screen acid-resistant and sensitive varieties, and has potential agricultural application value.

[0040] The specific examples described in the application are merely examples of the spirit of the present invention. Those skilled in the art may make various modifications or additions to the specific examples described in the present invention, or replace them in a similar manner, but they will not deviate from the spirit of the present invention or exceed the scope defined by the attached claims.

Claims

1. A method for quickly identifying the acid resistance of rapeseed varieties based on RGB images, characterized in that: The steps include: Step S1, dividing the rapeseed varieties to be screened into two treatment groups for potted planting, one group is placed in acidic soil, and the other group is placed in neutral soil, and other physical and chemical properties of the two groups of soils are kept consistent; Step S2, starting from the rapeseed bud stage, regularly using an RGB image acquisition device to simultaneously acquire top view and side view images until the end of the entire growth period; Step S3, segmenting the rapeseed plant regions in the top view and the side view by a deep learning algorithm, and extracting texture features and morphological features from the segmented images as feature values; Step S4, at each image acquisition time point, for all tested varieties, the characteristic value change rates between adjacent time points under acidic and neutral soil environments are calculated respectively, an acid stress discrimination model is established using a machine learning algorithm, key characteristic values ​​are screened and their weights are determined, and the acid resistance evaluation value of each variety at that time point is calculated based on the key characteristic value change rate and weight under the acidic soil environment; Step S5, based on the acid resistance evaluation value at each time point in the whole growth period obtained in step S4, analyze the importance of the acid resistance evaluation value at different time nodes to the overall evaluation, determine the time node weight, and calculate the comprehensive acid resistance evaluation value for the whole growth period in combination with the acid resistance evaluation value at each time point and its time node weight, and use the comprehensive acid resistance evaluation value for the whole growth period to judge the acid resistance of rapeseed varieties. The larger the comprehensive evaluation value, the stronger the acid resistance.

2. The method for rapidly identifying the acid resistance of rapeseed varieties based on RGB images according to claim 1, characterized in that: The image segmentation in step S3 is implemented in the following way: first, the original image is annotated to obtain a binary image label, and then the training set and the test set are divided according to a certain ratio. Then, the Segformer segmentation network is used to train the data set, and the optimal weight of the network is derived to segment all rapeseed images.

3. The method for rapid identification of acid resistance of rapeseed varieties based on RGB images according to claim 1, characterized in that: The texture features are extracted through grayscale histogram and grayscale co-occurrence matrix, and the obtained parameters include but are not limited to correlation, gradient advantage, energy, gradient mean, gradient entropy, differential moment, gradient standard deviation and third-order moment.

4. The method for rapidly identifying the acid resistance of rapeseed varieties based on RGB images according to claim 1, characterized in that: The morphological characteristics are calculated by processing the rapeseed binary image, and the obtained parameters include but are not limited to width, height, projected area, convex hull area, perimeter-to-area ratio, compactness and fractal dimension.

5. The method for rapid identification of acid resistance of rapeseed varieties based on RGB images according to claim 1, characterized in that: In step S4, the calculation formula of the characteristic value change rate is: in, It represents the rate of change of the eigenvalue with eigenvalue index f between adjacent time points t+1 and t; represents the feature value with index f extracted from the image collected at time point t+1, Indicates that the index extracted from the image collected at time point t is f eigenvalue, where F is the eigenvalue index set and contains all eigenvalues ​​used for calculation. .

6. The method for rapid identification of acid resistance of rapeseed varieties based on RGB images according to claim 5, characterized in that: In step S4, a random forest algorithm is used to establish an acid stress discrimination model, the input of which is the characteristic value change rate of the variety under the acidic soil environment and the neutral soil environment, and the output is the environment type; the model calculates the influence of each feature on the classification result through a feature importance measurement method, and uses it as the characteristic value weight w f Used for calculation of acid resistance evaluation value.

7. The method for rapidly identifying the acid resistance of rapeseed varieties based on RGB images according to claim 6, characterized in that: In step S4, the acid resistance evaluation value R at a single time point t The calculation formula is: 。 8. The method for rapidly identifying the acid resistance of rapeseed varieties based on RGB images according to claim 7, characterized in that: In step S5, the time node weight w t The calculation method is to perform principal component analysis on the acid resistance evaluation values ​​at different time points to obtain the principal component eigenvalues ​​and eigenvectors, where the principal component eigenvalues ​​represent the variance that the principal component can explain; then calculate the proportion of each principal component eigenvalue to the sum of all principal component eigenvalues, that is, the explained variance ratio; finally, weight the components of all principal component eigenvectors and the corresponding explained variance ratios to obtain the weight w at each time point. t .

9. The method for rapidly identifying the acid resistance of rapeseed varieties based on RGB images according to claim 8, characterized in that: In step S5, the comprehensive evaluation value of acid resistance during the whole growth period ARS total The calculation formula is: Among them, total indicates the total number of time points.

Images