Method for detecting early harm of tetranychus urticae based on Raman spectrum

Abstract

The invention provides a Raman spectrum-based method for detecting early harm of tetranychus urticae. The method comprises the following steps: preparing a detection sample: preparing a to-be-detected sample blade; acquiring a spectrum by using a portable Raman spectrometer; and determining whether the collected leaves are damaged by the tetranychus urticae or not based on the Raman spectrum. Reasonable utilization and conversion of carbon resources are accelerated due to the problem that carbon sequestration enzyme related information in an existing enzyme database is much and impure. The invention provides a sensitive, specialized, accurate and convenient detection method, a portable Raman spectrum is utilized to collect Raman spectrum information before and after the two-spotted spider mites harm the plant leaves, and rapid detection and identification of the two-spotted spider mites harm the plant leaves are realized.

Description

Technical Field

[0001] The invention belongs to the field of detection of spider mites that harm plants, and in particular to a method for detecting early-stage damage by two-spotted spider mites based on Raman spectroscopy. Background Art

[0002] The two-spotted spider mite (Tetranychus urticae), also known as the two-spotted spider mite, belongs to the genus Tetranychus of the family Tetranychidae and is a worldwide pest mite. It mainly gathers on the back of leaves to suck juice. After the juice is sucked, white spots will first appear on both sides of the main vein of the leaves. As the damage worsens, the leaves will turn grayish white, inhibiting the normal photosynthesis. As a result, it causes serious yield reductions in field crops, fruit trees and vegetables (SUN Yue-hua, ZHI Jun-rui, TIAN Tian, ​​et al. The damage of Tetranychus urticae on physiological index of kidney bean. Chinese Journal of Ecology, 2014, 31(5): 843-854). Among the resistant pests / mites that cause significant economic losses to crops worldwide, the two-spotted spider mite ranks fifth (Van Leeuwen T, Vontas J, Tsagkarakou A, Dermauw W, Tirry L. Acaricide resistance mechanisms in the two-spotted spider mite Tetranychus surticae and other important Acari: a review. Insect Biochem Mol Biol. 2010, 40(8): 563-572).

[0003] The two-spotted spider mite is particularly difficult to control due to its small size, fast reproduction rate, short generation cycle, and ability to reproduce both sexually and parthenogenetically. Through early detection, effective control measures can be taken when the density is low, which can reduce damage to plants, increase crop yield and quality, reduce control costs, and avoid environmental pollution and resistance problems caused by the large-scale use of pesticides. In addition, early detection can also help formulate scientific and reasonable integrated control strategies to protect the health of agricultural ecosystems. Therefore, early detection of two-spotted spider mites is crucial because it helps to detect and control pests in a timely manner and prevent their large-scale reproduction and spread. It can be seen that the detection of pests is a core component of integrated pest management (Leybourne DJ, Musa N, Yang P. Can artificial intelligence be integrated into the best monitoring schemes to help achieve sustainable agriculture? Anentomological, management and computational perspective. Agricultural and Forest Entomology. 2024, 1–10).

[0004] At present, there are three main types of detection methods for two-spotted spider mites. The first type is the visual detection method, which uses a magnifying glass or microscope to observe the damage of two-spotted spider mites. Although this method is simple and intuitive, it has many disadvantages, such as relying on the experience of the inspector, low efficiency, difficulty in early detection of two-spotted spider mites, being affected by environmental conditions, limited detection range and low accuracy. These shortcomings make it difficult to fully and accurately assess the actual damage of two-spotted spider mites, so it is usually necessary to combine other detection methods to improve the detection effect. The second type is the biochemical detection method, which uses PCR technology to detect two-spotted spider mite DNA in plant tissues, or detects changes in chlorophyll content in leaves, to achieve rapid and accurate detection and indirectly judge the damage of two-spotted spider mites. Although the biochemical detection method can provide valuable information in the detection of two-spotted spider mites, it also faces problems such as high cost, complex operation, high equipment requirements and possible environmental impact. The third type is the spectral information detection method, which uses spectral information to identify the characteristics of two-spotted spider mite damage for detection, which improves the efficiency and accuracy of detection, reduces labor costs, and supports more scientific prevention and control decisions.

[0005] In recent years, Raman spectroscopy has played an important role in pest and disease detection due to its rapid, non-destructive, water-resistant and micro-area resolution characteristics. Each pest has its own specific characteristic spectrum after damage. Based on the feasibility of using a portable Raman spectrometer to distinguish between healthy leaves and two-spotted spider mites, it can be used for early detection of two-spotted spider mites to prevent large-scale outbreaks on crops and affect crop yield and quality. Summary of the invention

[0006] The embodiment of the present invention provides a method for detecting early damage of two-spotted spider mites based on Raman spectroscopy, aiming to utilize Raman spectroscopy because of its rapidity, non-destructiveness, water resistance and micro-area resolution. Each pest has its own specific characteristic spectrum after damage. On the basis of using a portable Raman spectrometer to demonstrate the feasibility of Raman spectroscopy for differentiation, a rapid detection and identification method is developed using a Raman spectrometer and machine learning and other means to provide optional technical support and scientific guarantee for agricultural production.

[0007] A first aspect of an embodiment of the present invention provides a method for establishing an early damage detection model for two-spotted spider mites based on Raman spectroscopy, comprising the following steps:

[0008] (1) Preparation of test samples: Prepare the sample leaves to be tested, and pick the cotyledon leaves with different degrees of damage;

[0009] (2) Collecting spectra: Using a Raman spectrometer (e.g., a portable ATR3000 Raman spectrometer) to collect spectral data.

[0010] In specific operation, the parameters are set as follows: operating temperature -10~40℃, operating humidity, <95%, power supply voltage, DC 19V (+ / -5%), signal-to-noise ratio, 50μm, numerical aperture, 0.3; aperture, 7mm; center wavelength, 785nm (+ / -1nm); half-peak width, 0.08nm;

[0011] (3) Based on Raman spectroscopy: Determine whether the collected leaves are damaged by two-spotted spider mites. The damaged leaves have a spectral distribution of 1327 cm -1 The peak value at is relatively reduced, indicating that as the two-spotted spider mites damage the leaves, the chlorophyll content in the leaves decreases. When the leaves are severely damaged, most of the characteristic peaks can no longer be effectively detected.

[0012] In the above embodiment of the present invention, optionally, cotyledon leaves with different degrees of damage are picked.

[0013] In the above embodiment of the present invention, optionally, the cotyledon leaves with different degrees of damage include healthy leaves, slightly damaged leaves, moderately damaged leaves, and severely damaged leaves.

[0014] In the above embodiment of the present invention, five points with typical hazard characteristics are optionally marked on the blade, and data are collected twice at each of the marked points, with the front and back sides collected simultaneously. The file recording method is "repetition number-hazard number-method", such as 18-3-L is the Raman spectrum result of the hazard degree 3 of the 18th group of repetitions.

[0015] In the above embodiment of the present invention, optionally, the cotyledons and leaves with different degrees of damage are as follows: leaves that are uniformly green are considered healthy leaves, sporadic chlorosis and white dot-like symptoms appearing are considered mild damage, sporadic white dot-like damage symptoms that appear in blocks are considered moderate damage, and leaves that show large-scale chlorosis and damage symptoms that appear in patches are considered severe damage.

[0016] In the above embodiment of the present invention, optionally, the location of the typical damage feature is the traces of feeding by two-spotted spider mites appearing on leaves, that is, the degree of chlorosis of the leaves.

[0017] In the above embodiment of the present invention, optionally, the spectrum acquisition includes the steps of preprocessing and data processing the Raman spectrum data.

[0018] In the above embodiment of the present invention, optionally, the preprocessing is to first use the spectral data derived from the spectrometer and remove the instrument dark noise and baseline as the original Raman spectral data and convert it into a two-dimensional array format, use the Savitzky-Golay algorithm to smooth and reduce the noise of the data, then apply the standard normal variable change to standardize the data to eliminate the deviation between the spectra, and finally use the maximum-minimum normalization algorithm to achieve normalization of the spectral data.

[0019] In the above embodiment of the present invention, optionally, the data processing step is: using two machine learning classification algorithms, support vector machine and random forest algorithm, and convolutional neural network CNN algorithm in deep learning to realize the classification of the degree of damage to two-spotted spider mites on beans.

[0020] In the second aspect of the above embodiment of the present invention, the early damage detection method of spider mites based on Raman spectroscopy is to use the early damage detection model of spider mites based on Raman spectroscopy as described above for detection. Optionally, the specific method of detecting the damage of spider mites to beans is as follows:

[0021] (1) Classification model establishment: A portable Raman spectrometer is used to collect Raman spectral data of cotyledon leaves with different degrees of damage by two-spotted spider mites, and a machine learning model and a deep learning model are established to classify healthy leaves and damaged leaves. When the leaves are tested again later, the classification results of the classification model can be used to automatically determine whether the leaves are damaged by two-spotted spider mites;

[0022] (2) Damage detection: Leaf damage is classified based on the full-spectrum differences in Raman spectra between healthy leaves and damaged leaves. Through comparison, we finally selected the random forest machine learning classification algorithm for the classification of leaf two-spotted spider mite damage. The basis of random forest is the decision tree. The decision tree is a tree-structured classification model that divides data into different categories through a series of decision rules. Each node performs data segmentation based on a threshold of a certain feature until the leaf node gives the final classification result. Among them, in the feature selection process, random forest can help evaluate the importance of features and select the features that have the greatest impact on model performance. Therefore, random forest is widely used in data classification and image classification tasks, and is suitable for situations where data features are diverse and complex. The RF classification model is constructed based on the full-spectrum features of Raman data, which can extract the Raman spectrum intensity features and spectral curvature change features of each type of sample.

[0023] After the present invention adopts the above technical solution, it mainly has the following effects:

[0024] The two-spotted spider mite is particularly difficult to control due to its small size, fast reproduction rate, short generation cycle, and ability to reproduce both sexually and parthenogenetically. Therefore, early detection of the two-spotted spider mite is crucial because it helps to discover and control pests in a timely manner and prevent their large-scale reproduction and spread. The present invention first uses a portable Raman spectrometer to summarize and compare the chemical components represented by the characteristic peaks before and after the two-spotted spider mite feeding damage, and finds that the chemical structures that cause leaf chlorosis are different, based on which it can be distinguished whether there is spider mite damage. That is, it is verified that the use of a portable Raman spectrometer can detect whether spider mites are harmful or not. This is a step with practical application value and lays the foundation for the subsequent realization of instant field detection.

[0025] On this basis, using Raman spectrometer and machine learning methods, we explored the obvious differences between before and after the two-spotted spider mite feeding damage, and established a rapid differentiation model based on Raman spectroscopy. By combining Raman spectroscopy with machine learning algorithms for analysis, different labels were given before and after the two-spotted spider mite feeding damage. The model achieved classification and identification before and after the two-spotted spider mite feeding damage through comparative analysis of Raman spectra, and successfully distinguished before and after the two-spotted spider mite feeding damage. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required for use in the embodiments will be 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.

[0027] Figure 1The average spectrum and main characteristic peaks of the front side of the leaf before and after the two-spotted spider mite feeding damage in Example 1;

[0028] Figure 2 The average spectrum and main characteristic peaks of the reverse side of the leaf before and after the two-spotted spider mite feeding damage in Example 1;

[0029] Figure 3 The chlorophyll level and characteristic peak intensity value fitting curve of the front and back sides of leaves before and after the two-spotted spider mite feeding damage in Example 1;

[0030] Figure 4 The spectra and main characteristic peaks of ten sampling points on the front side of the two-spotted spider mite feeding damage of different degrees in Example 1;

[0031] Figure 5 The spectra and main characteristic peaks of ten sampling points on the back of the spider mite with different degrees of feeding damage in Example 1;

[0032] Figure 6 It is the Raman spectrum preprocessing operation process;

[0033] Figure 7 The Raman spectra after different preprocessing;

[0034] Figure 8 Comparison of classification accuracy of 1, 2 / 3 support vector machines in Example 2;

[0035] Fig. 9 The results of the 1, 2 / 3 support vector machine classification independent verification data set of Example 2;

[0036] Fig.10 Comparison of the classification accuracy of support vector machines 1 and 3 in Example 2;

[0037] Fig.11 The results of the independent validation data set for support vector machine classification 1 and 3 of Example 2;

[0038] Fig.12 Comparison of the classification accuracy of 1 and 2 / 3 random forests in Example 2;

[0039] Fig.13 The results of the 1 and 2 / 3 random forest classification independent validation datasets of Example 2;

[0040] Fig.14 Comparison of the random forest classification accuracy of 1 and 3 in Example 2;

[0041] Fig.15 The results of the 1 and 3 random forest classification independent validation data sets of Example 2;

[0042] Fig.16 This is a flow chart of the CNN model of Example 2;

[0043] Fig.17 This is the accuracy comparison of 1, 2 / 3 RF-CNN in Example 2;

[0044] Fig.18 The results of the 1, 2 / 3 RF-CNN independent validation datasets of Example 2;

[0045] Fig.19 Comparison of 1 and 3 RF-CNN accuracy in Example 2;

[0046] Fig. 20 These are the results of the 1 and 3 RF-CNN independent validation datasets of Example 2. DETAILED DESCRIPTION

[0047] In order to better understand the technical solution of the present invention, the embodiments of the present invention are described in detail below with reference to the accompanying drawings.

[0048] It should be clear that the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0049] The terms used in the embodiments of the present invention are only for the purpose of describing specific embodiments, and are not intended to limit the present invention. The singular forms "a", "said" and "the" used in the embodiments of the present invention and the appended claims are also intended to include plural forms, unless the context clearly indicates other meanings.

[0050] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the embodiments of the present invention are further described in detail below in conjunction with the drawings of the embodiments of the present invention, but the described embodiments are only some embodiments of the present invention, not all. Based on the embodiments of the present invention, other embodiments of the present invention that are not creative work by technicians in this field belong to the scope of protection of the present invention.

[0051] The present invention proposes a sensitive, specialized, accurate and convenient identification method, which uses portable Raman spectroscopy to collect Raman spectral information before and after the two-spotted spider mite damages plant leaves, thereby realizing rapid detection and identification of the two-spotted spider mite damages plant leaves.

[0052] The present invention firstly summarizes and compares the Raman spectral information and chemical information of the bean leaves before and after the two-spotted spider mite infests them by using a portable Raman spectrometer, and finds that the characteristic peak intensity is different. Then, the Raman spectrometer and machine learning methods are used to explore the differences between the two-spotted spider mite infestation on the bean leaves before and after, and a rapid differentiation model for bean spider mite damage detection based on Raman spectroscopy is established.

[0053] The method for detecting spider mites harming beans based on Raman spectroscopy comprises the following steps:

[0054] (1) Prepare the test sample: prepare the sample leaves to be tested;

[0055] (2) Collecting spectra: using a portable ATR3000 Raman spectrometer;

[0056] (3) Based on Raman spectroscopy: Determine whether the collected leaves are damaged by two-spotted spider mites.

[0057] 5 days after planting the bean seedlings, two-spotted spider mites were introduced into clean leaves (healthy leaves) for continuous damage. Leaves with different degrees of damage were selected according to different damage times. A total of 4 degrees of damage were selected, including healthy leaves, slightly damaged leaves, moderately damaged leaves, and severely damaged leaves. Leaves that were uniformly green were considered healthy leaves, sporadic chlorotic white dot symptoms were considered mild damage, sporadic white dot damage symptoms that appeared in blocks were considered moderate damage, and large areas of leaves turned green and the damage symptoms were concentrated into patches, which was considered severe damage.

[0058] Optionally, cotyledon leaves with different degrees of damage by two-spotted spider mites (healthy leaves, slightly damaged leaves, moderately damaged leaves, and severely damaged leaves) are picked.

[0059] Optionally, mark 5 locations with typical hazard characteristics on the leaf, collect data twice at each marked location, and collect data from both the front and back sides. The file recording method is "repetition number-hazard number-method", such as 18-3-L is the Raman spectrum result of the hazard degree 3 of the 18th group of repetitions.

[0060] Optionally, leaves that are uniformly green are considered healthy leaves, sporadic chlorosis with white dots appearing are considered mild damage, sporadic white dots that appear in blocks are considered moderate damage, and large areas of leaves that turn green and the damage appears in patches are considered severe damage.

[0061] Optionally, the locations of the typical damage features are as follows: traces of feeding by two-spotted spider mites appearing on leaves, and the degree of chlorosis of the leaves.

[0062] Optionally, the spectrum acquisition includes the steps of preprocessing and data processing the Raman spectrum data.

[0063] Optionally, the preprocessing is to first use the spectral data exported from the spectrometer to remove the instrument dark noise and baseline as the original Raman spectral data and convert it into a two-dimensional array format, use the Savitzky-Golay algorithm to smooth and reduce the noise of the data, then apply the standard normal variable variation (SVN) to standardize the data to eliminate the deviation between spectra, and finally use the maximum-minimum normalization algorithm to achieve normalization of the spectral data.

[0064] Optionally, the data processing step is: using two machine learning classification algorithms, support vector machine and random forest algorithm, and convolutional neural network (CNN) algorithm in deep learning to classify the degree of damage caused by two-spotted spider mites on beans.

[0065] Support Vector Machine (SVM) is a powerful classification and regression algorithm that classifies data by constructing an optimal hyperplane. The core idea of ​​SVM is to find an optimal hyperplane to maximize the interval between categories, thereby improving classification accuracy. SVM receives labeled training data, each sample consisting of a feature vector and a category label. With this data, the SVM algorithm maps the data into a high-dimensional space to find a hyperplane that can best distinguish different categories. In order to find the best hyperplane, SVM needs to solve a convex optimization problem. Through the Lagrange multiplier method, the dual problem is constructed, and the hyperplane parameters (w) and (b) are solved using an optimization algorithm such as the sequential minimal optimization algorithm (SMO). Through the optimization algorithm, SVM finds the optimal (w) and (b), ensuring that the hyperplane can maximize the interval between categories and enable the classifier to achieve good classification results on the training set. Support vector machines can effectively handle various classification tasks, whether it is linearly separable or nonlinearly separable data, and can provide accurate classification results. The power of SVM lies in its performance in high-dimensional feature space and its ability to process complex data through kernel functions.

[0066] Random Forest is an ensemble learning method that builds multiple classifiers based on decision trees and obtains the final classification result through a voting mechanism. It is mainly used for classification and regression tasks and is widely used in practical problems due to its high accuracy and robustness. Random Forest consists of multiple decision trees. It builds each decision tree by random sampling of data and random selection of features. The training data set for each decision tree is constructed by randomly extracting samples from the original data set. The training data for each tree is obtained by sampling with replacement, so the same data point may appear multiple times in a tree or not appear in a tree. When constructing each node of each decision tree, only a random subset of features is considered for splitting. Each tree is built independently during the training process, and each tree may have a different structure and feature selection. For classification tasks, Random Forest determines the final classification result by voting on the prediction results of all decision trees. Specifically, each tree votes for one sample, and the final classification result is the majority vote of all trees. By constructing multiple decision trees and combining their prediction results, Random Forest can effectively improve the accuracy and robustness of the model.

[0067] Optionally, the specific method for detecting the damage of beans caused by two-spotted spider mites is as follows: (1) Classification model establishment: Using a portable Raman spectrometer, collect Raman spectral data of cotyledon leaves with different degrees of damage from two-spotted spider mites (healthy leaves, leaves with mild damage, and leaves with moderate damage), establish machine learning models (support vector machine, random forest classification model) and deep learning models (RF-CNN), and classify healthy leaves and damaged leaves; when the leaves are tested again later, the classification results of the classification model can automatically determine whether the leaves are damaged by two-spotted spider mites. (2) Damage detection: Classify the damage to leaves based on the full spectral difference between the Raman spectra of healthy leaves and damaged leaves. Through comparison, we finally selected the random forest machine learning classification algorithm for the classification of leaf two-spotted spider mite damage. The basis of random forest is decision tree. Decision tree is a tree-structured classification model that divides data into different categories through a series of decision rules (nodes). Each node performs data segmentation based on the threshold of a certain feature until the leaf node gives the final classification result. In the feature selection process, random forest can help evaluate the importance of features and select the features that have the greatest impact on model performance. Therefore, random forests are widely used in data classification, image classification and other tasks, and are suitable for situations where data features are diverse and complex. We build an RF classification model based on the full spectrum features of Raman data, which can extract the Raman spectrum intensity features and spectral curvature change features of each type of sample.

[0068] The Raman spectra of cotyledon leaves with different degrees of damage from spider mites (healthy leaves, leaves with mild damage, and leaves with moderate damage) were collected using a portable Raman spectrometer, and machine learning models (support vector machine, random forest classification model) and deep learning models (RF-CNN) were established to classify healthy leaves and damaged leaves. Therefore, when the leaves are tested again later, the classification results of the classification model can automatically determine whether the leaves are damaged by spider mites.

[0069] Beneficial effects of this application:

[0070] The two-spotted spider mite is particularly difficult to control due to its small size, fast reproduction rate, short generation cycle, and ability to reproduce both sexually and parthenogenetically. Therefore, early detection of the two-spotted spider mite is crucial because it helps to discover and control the pest in a timely manner and prevent its large-scale reproduction and spread. This application first uses a portable Raman spectrometer to summarize and compare the chemical components represented by the characteristic peaks before and after the two-spotted spider mite feeding damage, and finds that the chemical structures that cause leaf chlorosis are different, based on which it can be distinguished whether there is spider mite damage. That is, it verifies that the use of a portable Raman spectrometer can be used to detect whether spider mites are harmful or not. This is a step with practical application value and lays the foundation for the subsequent realization of instant field detection.

[0071] On this basis, using Raman spectrometer and machine learning methods, we explored the obvious differences between before and after the two-spotted spider mite feeding damage, and established a rapid differentiation model based on Raman spectroscopy. By combining Raman spectroscopy with machine learning algorithms for analysis, different labels were given before and after the two-spotted spider mite feeding damage. The model achieved classification and identification before and after the two-spotted spider mite feeding damage through comparative analysis of Raman spectra, and successfully distinguished before and after the two-spotted spider mite feeding damage.

[0072] The present invention relates to the field of plant pest detection, and specifically to a method for detecting spider mites that damage beans based on Raman spectroscopy. The purpose is to solve the problem that the two-spotted spider mite is small in size, the symptoms of damage are not easy to detect, and the detection is difficult. The present application utilizes the characteristics of Raman spectroscopy that is fast, non-destructive, water-free, and can resolve micro-areas. The ATR3000 Raman spectrometer is used in combination with machine learning and deep learning algorithms to establish a Raman spectral classification model before and after the two-spotted spider mite damages the leaves of beans, and develop a rapid detection and identification method to provide optional technical support and scientific guarantee for agricultural production.

[0073] Example 1: Preliminary detection of plant leaves damaged by spider mites based on portable Raman spectrometer

[0074] Experimental materials and instruments.

[0075] The required bean plants and two-spotted spider mites were bred and raised by the predatory mite research group of the Biological Control Laboratory, Institute of Plant Protection, Chinese Academy of Agricultural Sciences; the portable Raman spectrometer ATR3000 was from the Changchun Institute of Optics, Fine Mechanics and Physics, Chinese Academy of Sciences.

[0076] The preliminary detection method of plant leaves damaged by two-spotted spider mites based on portable Raman spectrometer is as follows:

[0077] (1) Prepare the test sample: prepare the sample leaves to be tested;

[0078] Five days after the bean seedlings were planted, two-spotted spider mites were inoculated on clean leaves (healthy leaves) and continued to damage them. Leaves with different degrees of damage were selected according to different damage times. A total of four degrees of damage were selected, including healthy leaves, lightly damaged leaves, moderately damaged leaves, and severely damaged leaves. Leaves that were uniformly green were considered healthy leaves, sporadic chlorotic white dot symptoms were considered light damage, sporadic white dot damage symptoms that appeared in blocks were considered moderate damage, and large areas of leaves turned green and the damage symptoms were concentrated in pieces, considered severe damage.

[0079] (2) Find the hazard and select the characteristic parts for testing

[0080] Pick cotyledon leaves with different degrees of damage by two-spotted spider mites (healthy leaves, slightly damaged leaves, moderately damaged leaves, and severely damaged leaves). Mark 5 locations with typical damage characteristics on the leaves, and collect twice at each marked location, collecting from the front and back at the same time.

[0081] (3) Spectral information collection

[0082] ①Raman spectroscopy system

[0083] The ATR3000 portable Raman spectroscopy system was used for acquisition, and the parameter settings were as follows:

[0084] Table 1: Portable Raman spectroscopy system performance parameters

[0085]

[0086] ② The specific parameters used in the experiment are shown in Table 1

[0087] (4) Raman spectroscopy analysis

[0088] The Raman spectroscopy system detected the spectral information of healthy leaves, mildly damaged leaves, moderately damaged leaves, and severely damaged leaves, and used origin to obtain the average spectrum, see Figure 1 and Figure 2 . Figure 1 These are the average spectra and main characteristic peaks on the front side of leaves before and after two-spotted spider mite feeding. Figure 2 These are the average spectra and main characteristic peaks of the reverse side of leaves before and after two-spotted spider mite feeding.

[0089] from Figure 2 It can be seen that the overall morphology of the spectrum of healthy leaves and leaves damaged by two-spotted spider mites is similar. After analyzing the spectrum, it was found that the characteristic peaks appeared at 920, 1157, 1327, and 1527 cm -1 These characteristic peaks are attributed to hydroxyproline (919 cm -1 )、chlorophyll(1327cm -1 ), carotenoids (1157cm -1 ), carotenoids (1527cm -1 )

[0090] Table 2: Raman spectrum characteristic peaks and their attribution substances

[0091]

[0092] The spectra of healthy leaves and leaves damaged by two-spotted spider mites are generally similar, with higher characteristic peaks appearing at 750, 1157, 1327, and 1527 cm -1 The other main characteristic peaks are at 919, 1038, and 1443 cm-1 This is because the biomacromolecules in plants, such as nucleic acids, unsaturated fatty acids and carbohydrates, are similar in composition and structure. -1 Analysis shows that 1327cm -1 The peak intensity is higher, and the damaged blade is 1327cm -1 The peak at 1327cm is relatively reduced, indicating that as the two-spotted spider mite damages the leaves, the chlorophyll content in the leaves decreases. When the leaves are seriously damaged, most characteristic peaks can no longer be effectively detected because the leaf tissue composition changes. The chlorophyll level of the leaves was measured at the spectral detection point using a chlorophyll meter. -1 The peak intensity of chlorophyll characteristic peak is linearly fitted with chlorophyll content. -1 The peak intensity has a strong linear correlation (R 2 >95%), see Figure 3 The curves are the chlorophyll levels and characteristic peak intensity values ​​on the front and back of leaves before and after the two-spotted spider mite feeding. The Raman spectroscopy system detected 20 groups of spectral information of healthy leaves, mildly damaged leaves, moderately damaged leaves, and severely damaged leaves. Each group consists of three leaves: healthy leaves, mildly damaged leaves, and moderately damaged leaves. The front side of each leaf ( Figure 3 ), reverse ( Figure 4 ) collect ten Raman spectrum data each. Figure 4 These are the spectra and main characteristic peaks of ten sampling points with different degrees of damage from two-spotted spider mites.

[0093] This example verifies that the portable Raman spectrometer can distinguish the spectra of healthy leaves and leaves damaged by two-spotted spider mites by characterizing the chlorophyll level of leaves. It demonstrates that the portable Raman spectrometer can detect the spectra of healthy leaves and leaves damaged by two-spotted spider mites, laying the foundation for real-time detection in the field.

[0094] Example 2: Construction of a preliminary detection system for plant leaves damaged by spider mites based on a portable Raman spectrometer

[0095] Experimental materials and instruments

[0096] The required bean plants and two-spotted spider mites were bred and raised by the predatory mite research group of the Biological Control Laboratory, Institute of Plant Protection, Chinese Academy of Agricultural Sciences; the portable Raman spectrometer ATR3000 was from the Changchun Institute of Optics, Fine Mechanics and Physics, Chinese Academy of Sciences.

[0097] The preliminary detection method of plant leaves damaged by two-spotted spider mites based on portable Raman spectrometer is as follows:

[0098] (1) Prepare the test sample: prepare the sample leaves to be tested;

[0099] Five days after the bean plants were planted, two-spotted spider mites were inoculated into clean leaves (healthy leaves) for continuous damage. Leaves with different degrees of damage were selected according to different damage times. A total of three degrees of damage were selected, including healthy leaves, slightly damaged leaves, and moderately damaged leaves. Leaves that were uniformly green were considered healthy leaves, sporadic chlorotic white dot symptoms were considered slightly damaged, and sporadic white dot damage symptoms that aggregated into blocks were considered moderately damaged.

[0100] (2) Find the hazard and select the characteristic parts for testing

[0101] Pick cotyledon leaves with different degrees of damage by two-spotted spider mites (healthy leaves, slightly damaged leaves, moderately damaged leaves), mark 5 locations with typical damage characteristics on the leaves, collect twice at each marked location, and collect from both the front and back sides at the same time.

[0102] (3) Spectral information collection

[0103] ①Raman spectroscopy system

[0104] The ATR3000 portable Raman spectroscopy system was used for acquisition, and the parameter settings were as shown in Table 3:

[0105] Table 3: Portable Raman spectroscopy system performance parameters

[0106]

[0107]

[0108] ② The specific parameters used in the experiment are shown in Table 3

[0109] (4) Raman spectroscopy analysis

[0110] The Raman spectroscopy system detected the spectral information of healthy leaves, mildly damaged leaves, and moderately damaged leaves. There are 15 groups of Raman spectroscopy data, each of which consists of three leaves: healthy leaves, mildly damaged leaves, and moderately damaged leaves. Figure 3 ), reverse ( Figure 4 ) collect ten Raman spectrum data each. Figure 4 These are the spectra and main characteristic peaks of ten sampling points with different degrees of damage from two-spotted spider mites.

[0111] ① Raman spectroscopy data preprocessing

[0112] The original Raman spectrum of the leaves damaged by spider mites in the experiment contains interference information such as background solution signal, noise, fluorescence, etc. In order to facilitate the observation of Raman spectra and the comparison of Raman spectra between different samples, the spectrum needs to be preprocessed. The Raman spectrum preprocessing operation process is as follows: Figure 6:First, the spectral data exported from the spectrometer with the dark noise and baseline removed are used as the original Raman spectral data and converted into a two-dimensional array format. The Savitzky-Golay filtering algorithm is used to smooth and reduce the noise of the data. Then, the standard normal variable variation (SVN) is applied to standardize the data to eliminate the deviation between spectra. Finally, the maximum-minimum normalization algorithm is used to normalize the spectral data. The preprocessing effect is as follows Figure 7 , are Raman spectra after different preprocessing.

[0113] ②Raman spectroscopy data processing and analysis

[0114] A. Based on support vector machine classification model

[0115] The collected spectral data of healthy leaves, slightly damaged leaves, and moderately damaged leaves were divided into two groups, with healthy leaves in one group and slightly damaged leaves and moderately damaged leaves in another group. A support vector machine classification model was established using python to classify the Raman spectra of leaves according to whether they were damaged by spider mites. We will use the classification model to obtain the classification accuracy under different conditions by combining the spectral data after different preprocessing, and divide the training set into 80% and the test set into 20%. Figure 8 , which is the comparison of the classification accuracy of 1 and 2 / 3 support vector machines. The results of the validation data set are as follows Fig. 9 .

[0116] For the support vector machine classification model, preprocessing SG-SNV-NOR can improve the classification accuracy. The classification effect of the back of the leaf is significantly better than that of the front data, and the accuracy can be greater than 90%. Considering that the overlap between Class 2 leaves and Class 1 and Class 3 is high, the healthy leaves are divided into one group and the moderately damaged leaves into another group. The support vector machine model is used again for binary classification, with the training set divided into 80% and the test set divided into 20%. Fig.10 , which are the spectra and main characteristic peaks of ten sampling points on the back of the plant with different degrees of damage by two-spotted spider mites. The results of the verification data set are as follows Fig.11 .

[0117] Compared with the binary classifications 1, 2 / 3, the accuracy of the binary classifications 1 and 3 is significantly improved. The accuracy of the training set decreases after preprocessing, and the accuracy of the test set is obvious. After ten-fold cross-validation, the classification accuracy is still greater than 95%. The classification accuracy of the back spectrum is still higher than that of the front spectrum.

[0118] B. Based on random forest classification model

[0119] The collected spectral data of healthy leaves, slightly damaged leaves, and moderately damaged leaves are divided into two groups, with healthy leaves as one group and slightly damaged leaves and moderately damaged leaves as one group. We will use Python to establish a random forest classification model to classify the Raman spectra of leaves according to whether they are damaged by spider mites. We will use the classification model to obtain the classification accuracy under different conditions by combining the spectral data after different preprocessing, such as Fig.12 , which is the comparison of the classification accuracy of 1 and 2 / 3 random forests. The results of the validation dataset are as follows Fig.13 .

[0120] For the random forest classification model, the improvement of classification accuracy by preprocessing is small, but the classification effect is better than that of the support vector machine classification model. The classification accuracy of the spectral data on the back of the leaf can be greater than 95%, and the classification accuracy of the spectral data on the front can be greater than 90%.

[0121] Considering that the overlap between Class 2 leaves and Class 1 and 3 is high, the healthy leaves are divided into one group and the moderately damaged leaves into another group. The support vector machine model is used again for binary classification, with 80% of the training set and 20% of the test set. Fig. 9 , verify the data set results, such as Fig.10 , which is the comparison of classification accuracy of 1 and 3 support vector machines. Fig.11 These are the independent validation data set results for support vector machine classification 1 and 3 of Example 2. Fig.14 Comparison of the classification accuracy of random forests 1 and 3. Fig.15 Results of 1 and 3 random forest classification independent validation datasets.

[0122] Compared with the binary classification of 1, 2 / 3, the accuracy of the binary classification of 1 and 3 is significantly improved. The accuracy of the test set after filtering, standard normalization, and normalization preprocessing is significantly improved. After ten-fold cross validation, the classification accuracy is still greater than 95%. The classification accuracy of the back spectrum and the front spectrum is not much different. By comparison, the optimal classification combination is the SG-SNV-Nor-RF classification method, with an accuracy of 100%.

[0123] C. RF-CNN classification model based on neural network

[0124] At the same time, MATLAB software was used to establish the RF-CNN classification model, in which the spectral data was first reduced in dimension using random forest, the first 20 features were selected as the input for subsequent classification, and then a double-layer convolutional neural network was established to build the classification model. The CNN model flow chart is shown in the figure. Fig.16 . It includes the following parts:

[0125] Input layer: used to receive and process input data;

[0126] Fully connected layer: It fully connects each feature of the input data with 64 neurons. This means that each input feature is connected to each neuron, and each neuron has a weight and bias parameter;

[0127] ReLU (Rectified Linear Unit) activation function layer: It helps introduce nonlinearity, allowing the model to learn complex features;

[0128] Fully connected layer: It connects the output of the previous layer to 2 neurons to map the features to the final binary classification;

[0129] Softmax activation function: used for classification tasks, which can convert the output of the network into probability;

[0130] Loss layer: Calculates the cross entropy loss between the network's predictions and the actual labels. During training, the optimization algorithm adjusts the network parameters based on this loss to improve the accuracy of the model.

[0131] The collected spectral data of healthy leaves, slightly damaged leaves, and moderately damaged leaves are divided into two groups, one of which is healthy leaves, and the other is slightly damaged leaves and moderately damaged leaves. We will use the classification model to obtain the classification accuracy in different situations, such as Fig.17 , for the accuracy comparison of 1, 2 / 3 RF-CNN, and verify the data set results as follows Fig.18 .

[0132] Considering that the overlap between Class 2 leaves and Class 1 and 3 is high, the healthy leaves are divided into one group and the moderately damaged leaves into another group. The RF-CNN model is used again for binary classification. We will use the classification model to match the positive and negative spectral data to obtain the classification accuracy in different situations, such as Fig.19 , is the accuracy comparison of 1 and 3RF-CNN, and the results of the validation dataset are as follows Fig. 20 .

[0133] The accuracy of 1 and 3 classifications is significantly higher than that of 1 and 2 / 3 classifications. Compared with the machine learning classification method, the neural network classification method did not significantly improve the classification effect.

[0134] In summary, Example 1 verified the application potential of Raman spectroscopy in the detection of two-spotted spider mites in plants. Example 2 successfully constructed a classification system for distinguishing plant leaves damaged by two-spotted spider mites based on the portable Raman spectrometer ATR3000 by combining Raman spectroscopy with machine learning and neural network classification models.

[0135] Here, it is explained that in Example 2, data was collected in 4 cases, and the data was initially screened using 4 cases, but 3 cases were used in the subsequent cluster analysis, because once the "serious damage" data is introduced, 1-3 will be classified into one category, and 4 cases of serious damage will be classified into one category. That would not be the spider mite "early damage" detection model proposed in this application. Therefore, for this reason, the data used for the subsequent modeling are "healthy leaves, mildly damaged leaves, and moderately damaged leaves".

[0136] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A method for establishing an early damage detection model for two-spotted spider mites based on Raman spectroscopy, characterized in that: The following steps are involved: (1) Preparation of test samples: Prepare the sample leaves to be tested and pick the cotyledon leaves with different degrees of damage; (2) Collecting spectra: Collecting spectral data using a Raman spectrometer (e.g., a portable ATR3000 Raman spectrometer); (3) Based on Raman spectroscopy: Determine whether the collected leaves are damaged by two-spotted spider mites. The damaged leaves have a spectral distribution of 1327 cm -1 The peak value at is relatively reduced, indicating that as the two-spotted spider mites damage the leaves, the chlorophyll content in the leaves decreases. When the leaves are severely damaged, most of the characteristic peaks can no longer be effectively detected.

2. The method for detecting early damage of spider mites based on Raman spectroscopy according to claim 1, characterized in that: Choose leaves with different degrees of damage according to different damage times.

3. The method for establishing an early damage detection model for Tetranychus urticae based on Raman spectroscopy according to claim 2, characterized in that: The cotyledon leaves with different degrees of damage include healthy leaves, slightly damaged leaves, moderately damaged leaves, and severely damaged leaves.

4. The method for establishing an early damage detection model for Tetranychus urticae based on Raman spectroscopy according to claim 3, characterized in that: Mark 5 locations with typical damage characteristics on the leaves, collect twice at each marked location, and collect from both the front and back sides; The file recording method is "repetition number-hazard number-method", and 18-3-L is the Raman spectrum result of the hazard level 3 of the 18th group of repetitions.

5. The method for establishing an early damage detection model for Tetranychus urticae based on Raman spectroscopy according to claim 3, characterized in that: The cotyledons and leaves with different degrees of damage are as follows: leaves that are uniformly green are considered healthy leaves, sporadic chlorosis and white dot-like symptoms appear as mild damage, sporadic white dot-like damage symptoms that appear in blocks are considered moderate damage, and leaves that have large areas of chlorosis and damage symptoms that appear in patches are considered severe damage.

6. The method for establishing an early damage detection model for Tetranychus urticae based on Raman spectroscopy according to claim 4, characterized in that: The typical damage characteristics are the presence of traces of feeding by two-spotted spider mites on leaves, that is, the degree of leaf chlorosis.

7. The method for establishing an early damage detection model for Tetranychus urticae based on Raman spectroscopy according to any one of claims 1 to 6, characterized in that: The spectrum acquisition includes the steps of preprocessing and data processing the Raman spectrum data.

8. The method for establishing an early damage detection model for Tetranychus urticae based on Raman spectroscopy according to claim 7, characterized in that: The preprocessing is to first remove the instrument dark noise and baseline from the spectral data exported from the spectrometer as the original Raman spectral data and convert it into a two-dimensional array format, use the Savitzky-Golay algorithm to smooth and reduce noise on the data, then apply the standard normal variable variation to standardize the data to eliminate the deviation between spectra, and finally use the maximum-minimum normalization algorithm to normalize the spectral data.

9. The method for establishing an early damage detection model for Tetranychus urticae based on Raman spectroscopy according to claim 7, characterized in that: The data processing steps are: using two machine learning classification algorithms, namely support vector machine and random forest algorithm, and convolutional neural network (CNN) algorithm in deep learning to classify the degree of damage to two-spotted spider mites on beans.

10. A method for detecting early damage of spider mites based on Raman spectroscopy, characterized in that: The specific method of detecting the early damage of spider mites by two-spotted spider mites by using the Raman spectroscopy-based early damage detection model as claimed in claim 8 or 9 is as follows: (1) Establishment of classification model: A portable Raman spectrometer is used to collect Raman spectral data of cotyledon leaves with different degrees of damage caused by two-spotted spider mites, and a machine learning model and a deep learning model are established to classify healthy leaves and damaged leaves. When the leaves are tested again later, the classification results of the classification model can be used to automatically determine whether the leaves are damaged by two-spotted spider mites. (2) Damage detection: Classify the damage of leaves based on the full-spectral difference of Raman spectra between healthy leaves and damaged leaves; use the random forest machine learning classification algorithm to classify the damage of leaf two-spotted spider mites; the basis of random forest is decision tree; A decision tree is a tree-structured classification model that divides data into different categories through a series of decision rules; each node performs data segmentation based on a threshold of a certain feature until the leaf node gives the final classification result.

Images