Cotton aphid condition evaluation method based on hyperspectral technology

Abstract

The invention discloses a cotton aphid condition evaluation method based on a hyperspectral technology, which comprises the following steps: dividing a cotton test field into a destructive sampling area and a non-destructive sampling area, and sampling; measuring cotton aphid quantity data, cotton hyperspectral data after aphid damage, and cotton growth index and physiological index data; analyzing the obtained data to obtain a spatial and temporal distribution rule of the number of cotton aphids, change characteristics of hyperspectral reflectivity of cotton after aphid damage, and change characteristics of growth indexes and physiological indexes of cotton; and processing the obtained hyperspectral data of the cotton subjected to aphid damage, screening a spectrum sensitive wave band, and constructing and verifying a cotton aphid quantity monitoring model. According to the evaluation method, a scientific theoretical basis and method are provided for obtaining cotton field insect pest situation information, a new thought is provided for scientific management of cotton insect pests, cotton aphid damage can be effectively relieved, sustainable development of the cotton industry is guaranteed, and the goals of high quality and high yield are achieved.

Description

Technical Field

[0001] The invention belongs to the technical field of cotton pest evaluation, and more specifically, the invention relates to a method for evaluating cotton aphids based on hyperspectral technology. Background Art

[0002] Cotton field pests have always affected the healthy production of cotton and restricted the overall revitalization of my country's cotton industry. Cotton aphids are one of the main pests that harm cotton. They mainly attach to the back of cotton leaves and young stems, suck sap, cause leaf curling, plant dwarfing, etc., affect photosynthesis and the development of nutritional organs, can spread a variety of plant viruses, and reduce the quality grade of cotton. Cotton aphids grow and develop very quickly. Under a suitable temperature environment, they are prone to explosive growth. At present, the acquisition of cotton aphid information is mainly manual investigation, which is time-consuming and labor-intensive, and seriously affects the prevention and control of cotton aphids. Therefore, timely acquisition of cotton aphid pest information is an effective guarantee for its comprehensive prevention and control. Hyperspectral remote sensing monitoring technology has the advantages of real-time, rapid, and non-destructive monitoring. It has been widely used in crop identification and planting area extraction, yield estimation, nutrient detection, etc., which has greatly promoted the development of agricultural informatization. By analyzing the changing patterns of crop canopy and single leaf spectral information, it can be judged whether the crop is eroded by pests and the extent of pests. Therefore, based on the remote and non-destructive advantages of spectral technology, accurate monitoring of cotton aphid information in cotton fields is of great significance for preventing and controlling aphid infestations and blocking the worsening of aphid infestations. Summary of the invention

[0003] An object of the present invention is to solve at least the above problems and / or disadvantages and to provide at least the advantages which will be described hereinafter.

[0004] In order to achieve these purposes and other advantages of the present invention, a method for evaluating cotton aphids based on hyperspectral technology is provided, comprising the following steps:

[0005] Step 1: Divide the cotton test site into a destructive sampling area and a non-destructive sampling area for sampling;

[0006] Step 2: Determine the number data of cotton aphids, the hyperspectral data of cotton after aphid infestation, the growth index of cotton, and the physiological index data;

[0007] Step 3: Analyze the data obtained in step 2 to obtain the temporal and spatial distribution of the number of cotton aphids, the change characteristics of the hyperspectral reflectance of cotton after aphid infestation, and the change characteristics of cotton growth indicators and physiological indicators;

[0008] Step 4: Process the hyperspectral data of cotton after aphid infestation obtained in step 2, screen the spectral sensitive bands, and build and verify the cotton aphid population monitoring model.

[0009] Preferably, in step one, the sampling method for the destructive sampling area is random sampling, and each time sampling is performed, 5 cotton plants with uniform growth are randomly selected from each variety in this area; the sampling method for the non-destructive sampling area is to randomly select 4 to 6 fixed points for each variety in this area, and each fixed point selects 3 to 5 consecutive cotton plants with uniform growth as continuous survey points, and marks them with signs.

[0010] Preferably, in step 2, the specific method for determining the number data of cotton aphids is:

[0011] S21. Destructive sampling area: The cotton plants were divided into three parts: upper layer, middle layer and lower layer according to plant height; the main stem leaves of each plant were removed, the first main stem leaf at the bottom was the inverted leaf, and the remaining main stem leaves were marked from bottom to top; the total number of cotton aphids on each main stem leaf, as well as the total number of cotton aphids on the upper leaves, middle leaves, lower leaves and upper stems, middle stems and lower stems were counted;

[0012] S22. Non-destructive sampling area: Investigate the total number of cotton aphids on each cotton plant at each designated point and convert it into the number of aphids per 100 plants.

[0013] Preferably, in the step 2, the specific method for determining the hyperspectral data of cotton after aphid infestation is: using a PSR-1100f portable ground object spectrometer for spectral acquisition, the spectral measurement range is 320-1100nm, the probe has a field of view of 25°, the resolution is 3.2nm, the canopy spectrum is scanned more than 10 times on average, and white board correction is required every 10-15 minutes during spectral data acquisition, the white board is placed flat, and the surface of the white board cannot be blocked by shadows during correction;

[0014] Destructive sampling area: collect the spectral reflectance of each main stem leaf, avoid leaf veins when collecting data, divide each leaf into three parts: upper, left and right, collect one spectral curve for each part, and perform whiteboard correction before each spectrum collection and measurement; Non-destructive sampling area: collect canopy spectra of each fixed-point cotton plant;

[0015] All spectral measurements were performed from a height of 50 cm above the cotton plants, and spectral data were acquired under clear skies with minimal or no wind between 11:00 and 14:00.

[0016] Preferably, in step 2, the cotton growth indicators include: plant height and leaf area; the cotton physiological indicators include: relative chlorophyll content, single leaf moisture content, anthocyanin content, malondialdehyde content, peroxidase activity and catalase activity.

[0017] Preferably, in step 4, the specific method for processing the hyperspectral data of cotton after aphid infestation is:

[0018] S41, hyperspectral data preprocessing: spectral data preprocessing using convolution smoothing;

[0019] S42. Spectral data transformation: The spectral data is processed by using first-order differential, second-order differential, logarithmic and inverse transformations.

[0020] Preferably, in step 4, the specific method for screening the spectral sensitive bands is: using the Pearson correlation coefficient method and the continuous projection method to screen the characteristic bands related to the number of cotton aphids.

[0021] Preferably, in step 4, the specific method for constructing and verifying the cotton aphid quantity monitoring model is:

[0022] Construct a cotton aphid population monitoring model: linear regression, extreme gradient boosting tree and partial least squares regression were used to construct a single leaf cotton aphid population monitoring model and a stratified cotton aphid population monitoring model;

[0023] Sample set division and model accuracy verification: The RAND function was used to generate random numbers, and the collected samples were randomly sorted and divided into training set: verification set = 2:1. Three indicators, namely, determination coefficient, root mean square error and relative analytical error, were selected to analyze the model accuracy.

[0024] Preferably, the calculation formula of the first-order differential is:

[0025]

[0026] The calculation formula of the second-order differential is:

[0027]

[0028] In the formula, R' λ(i) is the first-order differential value at wavelength i, R λ(i+1) and R λ(i-1) are the spectral reflectance values ​​at wavelengths i+1 and i-1, respectively, R” λ(i) is the second-order differential at wavelength i, R' λ(i+1) and R' λ(i-1) are the first-order differential values ​​at wavelengths i+1 and i-1 respectively, and 2Δλ represents the differential interval.

[0029] Preferably, the calculation formula of the logarithm is:

[0030] R(i)=ln(r(i));

[0031] In the formula, i represents the number of bands, r(i) represents the i-th spectral reflectance, and R(i) represents the i-th spectral reflectance after logarithmic transformation, where i = 400-1000;

[0032] The calculation formula of the reciprocal is:

[0033]

[0034] Wherein, i represents the number of bands, r(i) represents the i-th spectral reflectance, and R(i) represents the i-th spectral reflectance after reciprocal transformation, where i=400~1000.

[0035] The present invention includes at least the following beneficial effects: the cotton aphid pest evaluation method based on hyperspectral technology of the present invention investigates the pest information of cotton aphids, clarifies the temporal and spatial distribution patterns of cotton aphids, determines the optimal monitoring layer / leaf position for the number of cotton aphids, analyzes the change characteristics of the hyperspectral reflectance of a single leaf / canopy of cotton after cotton aphid infestation, explores the impact of cotton aphid infestation on cotton growth indicators and physiological indicators, constructs a cotton aphid quantity monitoring model based on hyperspectral technology, explores a fast and accurate cotton aphid pest information monitoring method, provides a scientific theoretical basis and method for obtaining cotton field pest information, provides a new idea for the scientific management of cotton pests, and can effectively reduce cotton aphid damage through precise monitoring and timely prevention and control, ensure the sustainable development of the cotton industry, and achieve the goal of high quality and high yield.

[0036] The cotton aphid situation evaluation method based on the hyperspectral technology provided by the present invention was used to evaluate the cotton aphid situation of Xinluzhong No. 81, Tahe No. 2 and J8031 cotton, and the results were as follows:

[0037] (1) There are two peaks of cotton aphids during the cotton growth period. The first peak is 54-67 days after sowing, and the second peak is 102 days after sowing. The number of cotton aphids is lower than before.

[0038] (2) When the aphid damage level increased, the anthocyanin content of Tahe No. 2 and J8031 gradually decreased; the malondialdehyde content showed a trend of first decreasing and then increasing with the increase of aphid damage level; the peroxidase activity showed a "V"-shaped change with the increase of damage; the catalase activity gradually increased with the increase of damage;

[0039] (3) The spectral reflectance of a single leaf is in the range of 757-1000nm, the reflectance of the aphid damage level is 2>3>0>1, and there is no obvious change in the position of the red edge. By comparing the model evaluation results, it is concluded that the FD+Pearson model based on the extreme gradient boosting tree is the most stable. 2 =0.79, RMSE = 53.50 head / leaf, RPD = 1.43, validation set R 2 =0.71, RMSE = 41.71 heads / leaf, RPD = 1.48, which can well predict the number of cotton aphids;

[0040] (4) As the vertical structure of the canopy moves downward, the spectral reflectance gradually decreases, and the number of cotton aphids also gradually decreases; in the range of 680-750nm, the red edge position shifts to the blue as the aphid damage deepens; by comparing the model evaluation results, it is concluded that the RG+SPA model based on the extreme gradient boosting tree is the most stable, and the training set R 2 =0.98, RMSE = 128.41 head / layer, RPD = 3.96, validation set R 2 =0.72, RMSE=231.84 heads / layer, RPD=1.52, which can well predict the number of cotton aphids.

[0041] Other advantages, objectives and features of the present invention will be embodied in part through the following description, and in part will be understood by those skilled in the art through study and practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] Figure 1 It is the technical roadmap of the present invention;

[0043] Figure 2 It is a schematic diagram of the test site division of the present invention;

[0044] Figure 3 It is a schematic diagram of cotton plant division of the present invention;

[0045] Figure 4 The variation trend of the number of cotton aphids on cotton plants of different varieties over time in the embodiment;

[0046] Figure 5 The number of cotton aphids per hundred plants on cotton plants of different varieties varies with time in the examples;

[0047] Figure 6 The effect of the number of cotton aphids on the height of cotton plants in the embodiment;

[0048] Figure 7 (A) is the effect of different aphid damage levels on the anthocyanin content of cotton. Figure 7 (B) Effects of different aphid damage levels on cotton MDA content; Figure 7 (C) Effects of different aphid damage levels on cotton POD activity; Figure 7 (D) Effects of different aphid damage levels on cotton CAT activity;

[0049] Figure 8 (A) is the correlation between different leaf positions and the number of cotton aphids per plant; Figure 8 (B) is the correlation between the number of cotton aphids per plant and the number of cotton aphids in each part; Figure 8 (C) is the correlation between the number of cotton aphids per 100 plants and the number of cotton aphids in each part;

[0050] Fig. 9is the reflectivity of cotton leaves under different numbers of cotton aphids in the embodiment;

[0051] Fig.10 The spectral variation characteristics of single leaves with different pretreatments in the embodiment;

[0052] Fig.11 The correlation between the single leaf spectrum transformation and the number of cotton aphids in the embodiment;

[0053] Fig.12 is the cotton layer reflectivity under different cotton aphid numbers in the embodiment;

[0054] Fig.13 The spectral variation characteristics of different pre-processing layers in the embodiment;

[0055] Fig.14 is the correlation between the hierarchical spectrum transformation and the number of cotton aphids in the embodiment;

[0056] Fig.15 It is the verification of the single leaf cotton aphid number estimation model based on LR in the embodiment;

[0057] Fig.16 It is the verification of the single-leaf cotton aphid number estimation model based on XGBoost in the embodiment;

[0058] Fig.17 It is the validation of the single leaf cotton aphid number estimation model based on PLSR in the embodiment;

[0059] Fig.18 It is the verification of the LR-based stratified cotton aphid quantity estimation model in the embodiment;

[0060] Fig.19 Verification of the hierarchical cotton aphid quantity estimation model based on XGBoost in the embodiment;

[0061] Fig. 20 This is the validation of the PLSR-based hierarchical cotton aphid population estimation model in the example. DETAILED DESCRIPTION

[0062] The present invention is further described in detail below in conjunction with the accompanying drawings so that those skilled in the art can implement the invention with reference to the description.

[0063] Example

[0064] A method for evaluating cotton aphid situation based on hyperspectral technology, the technical route is as follows Figure 1 As shown, including:

[0065] 1. Overview of the test site

[0066] The test was conducted in Aral City, Xinjiang in 2022-2023. The area is located in the mid-latitude zone of the northern hemisphere and belongs to the warm temperate continental arid desert climate zone. It has scarce precipitation, low moisture content in the air, and good atmospheric transparency. The average annual solar radiation in the area is 559.4-612.1KJ·cm, the sunshine hours are about 2996h / year, the sunshine is 100%, the annual accumulated temperature ≥10℃ is above 4000℃, the frost-free period is 18-224d, and the average annual temperature is 10.8℃. The rainfall is scarce, with an average annual precipitation of 50mm, an average annual evaporation of 1976.6-2558.9mm, and the groundwater depth is less than 3 meters. The surface evaporation is strong and the air is very dry. It is a typical warm temperate continental arid desert climate and a typical irrigated agricultural area. The rainfall in the experimental site in 2022 is mainly concentrated in 98-125 days after sowing, the historical highest temperature is 95 days after sowing, and the lowest temperature is 11 days after sowing; the rainfall in 2023 is mainly concentrated in 78-112 days after sowing, the historical highest temperature is 95 days after sowing, and the lowest temperature is 22 days after sowing.

[0067] 2. Experimental Design

[0068] The tested varieties were Xinluzhong 81, Tahe 2, and J8031, and the sowing dates for the two years were April 25 and April 14, respectively. Figure 2 As shown in the figure, the test site is divided into a destructive sampling area and a non-destructive sampling area. The field survey in the destructive sampling area is random sampling. Each time sampling, 5 cotton plants with uniform growth are randomly selected from each variety in this area, and the number of cotton aphids on each main stem leaf and the remaining leaves and stems is counted, and the spectral data of each main stem leaf is measured. In the non-destructive sampling part, 4-6 fixed points are randomly selected from each variety in the area, and 3-5 consecutive cotton plants with uniform growth are selected at the fixed points, and marked with signs to facilitate subsequent investigations. Insect information is investigated every 7-10 days, and the spectral data of single leaves and canopies are measured. The test site was not treated with aphid prevention agents, and was irrigated 10 times during the entire growth period. All other production management was carried out according to field production. If the weather is bad when collecting various data, the tests will be postponed and records will be kept.

[0069] 3. Determination indicators

[0070] (1) Survey on the number of cotton aphids

[0071] Destructive sampling: The cotton plants were divided into three parts according to their leaf positions and layers, namely the upper layer (C1), the middle layer (C2), and the lower layer (C3). The main stem leaf (L0) of each plant was removed, and the first main stem leaf at the bottom was the inverted leaf. The remaining main stem leaves were marked from bottom to top, such as Figure 3The total number of cotton aphids on each main stem leaf was counted, and the total number of cotton aphids on the upper leaves (L1), middle leaves (L2), lower leaves (L3) and upper stems (S1), middle stems (S2), and lower stems (S3) were counted respectively.

[0072] Non-destructive sampling: 4-6 fixed points were selected for each of the three varieties. 3-5 consecutive cotton plants with uniform growth were selected as continuous survey points. The total number of cotton aphids on each plant was surveyed each time sampling was carried out, and finally converted into the number of aphids per 100 plants (H0).

[0073] (2) Cotton hyperspectral data measurement after aphid infestation

[0074] This experiment uses the PSR-1100f portable ground spectrometer from Spectral Evolution, USA, for spectral collection. The spectral measurement range is 320-1100nm (ultraviolet-visible-near infrared), the probe has a 25° field of view, a resolution of 3.2nm, and the canopy spectrum is averaged over 10 scans. During spectral data collection, whiteboard correction is required every 10-15 minutes. The whiteboard should be placed flat and the surface of the whiteboard should not be blocked by shadows during correction.

[0075] For cotton randomly obtained in the destructive sampling area, the spectral reflectance of each layer is collected in turn using a portable spectrometer; all main stem leaves are cut off from the main stem of the cotton plant, and the spectral reflectance of each main stem leaf is collected. When collecting data, avoid the leaf veins, divide each leaf into three parts: upper, left, and right. One spectral curve is collected from each part, and white board calibration is required before each spectrum is collected and measured;

[0076] The canopy spectra of cotton plants at each fixed point were collected in the non-destructive sampling area.

[0077] All spectral measurements were performed from a height of 50 cm above the cotton plants, and spectral data were acquired between 11:00 and 14:00 (Beijing Time) under a clear sky with minimal or no wind.

[0078] (3) Growth indicators

[0079] Plant height: Use a tape measure to measure from the cotyledon node to the growing point (after topping, it is from the cotyledon node to the top fruit branch node).

[0080] Leaf area: Use LI-3100C desktop leaf area meter to scan leaf area, keep the leaf surface clean and flat to ensure the measured data is accurate.

[0081] (4) Physiological indicators

[0082] Relative chlorophyll content (SPAD) value: Use SPAD-502 chlorophyll meter, select the main stem leaves of each plant, wipe the leaves clean and then start the measurement.

[0083] Single leaf moisture content: All main stem leaves of each plant were collected, weighed and placed in marked envelopes and taken back to the laboratory. They were incubated at 105°C for 30 minutes and then dried at 80°C to constant weight.

[0084] Anthocyanins: The anthocyanin content was determined by the acidified methanol method.

[0085] Malondialdehyde (MDA): Thiobarbituric acid (TBA) was used to determine the MDA content.

[0086] Peroxidase (POD): measured using a peroxidase content detection kit (Nanjing Jiancheng).

[0087] Catalase (CAT): Catalase content detection kit (Nanjing Jiancheng) was used for determination.

[0088] 4. Spectral data processing

[0089] (1) Spectral data preprocessing

[0090] In the process of acquiring hyperspectral data, the reflectivity will generate noise due to environmental interference. This noise cannot be avoided in the data collection process. This experiment uses convolution smoothing (Savitzky-Golay, SG) for spectral data preprocessing. Due to the influence of the experimental instrument and dark current, noise will be generated at the front and back ends of the spectrum. Therefore, the obvious noise band is eliminated, and the hyperspectral data in the range of 400-1000nm is selected for modeling analysis.

[0091] (2) Spectral data transformation

[0092] Transforming spectral data can reduce the noise generated by environmental interference when collecting data, and can improve the sensitivity of the original spectrum (Reflection Spectrum Without Pretreatment Method, RS) and the optimization of characteristic bands. This experiment uses the transformation of first-order derivative (FD), second-order derivative (SD), logarithm (LG), and reciprocal (RG) to process spectral data. The calculation formula is as follows:

[0093] First-order differential: Second order differential:

[0094] In the formula, R' λ(i) is the first-order differential value at wavelength i, R λ(i+1) and Rλ(i-1) are the spectral reflectance values ​​at wavelengths i+1 and i-1, respectively, R” λ(i) is the second-order differential at wavelength i, R' λ(i+1) and R' λ(i-1) are the first-order differential values ​​at wavelengths i+1 and i-1, respectively, and 2Δλ represents the differential interval;

[0095] reciprocal:

[0096] Where i represents the number of bands, r(i) represents the i-th spectral reflectance, and R(i) represents the i-th spectral reflectance after inverse transformation, where i = 400, 401, ... 1000;

[0097] Logarithm: R(i)=ln(r(i));

[0098] Where i represents the number of bands, r(i) represents the i-th spectral reflectance, and R(i) represents the i-th spectral reflectance after logarithmic transformation, where i = 400, 401, … 1000.

[0099] 5. Spectral sensitive band screening

[0100] Since there is a large amount of redundant data and collinearity problems in the spectral data of the entire band that affect the accuracy, in order to reduce the complexity of the model and screen out sensitive bands to reduce the data dimension, this experiment used the Pearson Correlation Coefficient method (Pearson) and the Successive Projections Algorithm (SPA) to screen the characteristic bands with strong correlation with the number of cotton aphids. The Pearson correlation coefficient method selected the 10 bands with the highest correlation for subsequent modeling and analysis.

[0101] (1) Pearson correlation coefficient method (Pearson)

[0102] The correlation coefficient is an indicator that reflects the closeness of the correlation between corresponding variables. The Pearson correlation coefficient calculation formula is as follows:

[0103]

[0104] Where i = 1, 2, ..., n, n is the total number of samples of spectral data, are the means of the ith spectral absorption value and the physicochemical value, x i ,y i are the spectral absorption value and sample physicochemical value at the i-th wavelength respectively.

[0105] The correlation coefficient ρ between two vectors x and y of length n is a scalar, represented by the absolute value of the Pearson correlation coefficient. The value of ρ is between 0 and 1. Generally speaking, the closer ρ is to 1, the stronger the correlation between the two vectors. Conversely, the weaker the correlation between the two quantities.

[0106] (2) Continuous Projection Method (SPA)

[0107] For specific steps and principles, refer to Wang Qianlong, Niu Fangpeng et al. (Wang Qianlong. Research on soil total nitrogen prediction modeling based on soil visible-near infrared spectral library [D]. Zhejiang University, 2015) and (Niu Fangpeng, Li Xinguo, Maimaiti Tuersun Aizezi, et al. Hyperspectral estimation of soil organic carbon content in the lakeside oasis on the west bank of Bosten Lake based on continuous projection algorithm [J]. Journal of Zhejiang University (Agriculture and Life Sciences Edition), 2021, 47(05): 673-682.).

[0108] 6. Model construction and verification

[0109] (1) Linear regression (LR); (2) Extreme Gradient Boosting Tree (XGBoost); (3) Partial Least Squares Regression (PLSR);

[0110] (4) Sample set division and model accuracy verification

[0111] The RAND function was used to generate random numbers, and the collected samples were randomly sorted and divided into training set and validation set in a ratio of 2:1. 2 ), Root Mean Square Error (RMSE) and Relative Percent Deviation (RPD) are used to analyze the model accuracy. 2 The closer it is to 1, the smaller the RMSE value is. When RPD is closer to 2, the model is more stable, the accuracy is higher, and the model is better. The calculation formula is as follows:

[0112]

[0113] Where i represents the i-th sample data, x i is the measured value of the i-th sample data, is the average value of the measured values, y i is the predicted value of the i-th sample data, is the average of the predicted values.

[0114] Results and Analysis

[0115] 1. Temporal and spatial distribution of cotton aphids

[0116] 1.1 Distribution characteristics of cotton aphid numbers on main stem leaves over time

[0117] 2022: In the early stage, cotton aphids were mainly concentrated in the position of leaves 2 to 4; as time went on, the number of cotton aphids decreased rapidly, and 71 days after sowing, they were mainly concentrated in leaves 5 to 9. Since 78 days after sowing, the number of cotton aphids began to decrease rapidly, and there was no obvious change in the distribution of each leaf position.

[0118] 2023: There are two peak periods of cotton aphids. Tahe No. 2 reached the first peak of cotton aphids 54 days after sowing, with cotton aphids mainly concentrated in the 6th to 10th leaves, and the maximum number of a single leaf reached 360. The second peak of cotton aphids 102 days after sowing was reached, mainly concentrated in the 2nd to 5th leaves, and the maximum number of a single leaf reached 227. J8031 reached the first peak of cotton aphids 54 days after sowing, with cotton aphids located in the 3rd to 7th leaves, and the maximum number of a single leaf reached 285. The second peak of cotton aphids 89 days after sowing was reached, mainly concentrated in the 6th to 10th leaves, and the maximum number of a single leaf reached 141.

[0119] From the above, we can see that cotton aphids mainly attach to the leaves in the middle and upper parts of the main stem.

[0120] 1.2 Vertical distribution characteristics of cotton aphids on cotton plants

[0121] 2022: In the early stage, cotton aphids gathered more on the upper part of the cotton plants of Xinluzhong No. 81, and the order of proportion on the cotton plants was main stems and leaves > upper leaves > upper stems > middle leaves > middle stems > lower leaves > lower stems. The distribution of cotton aphids in Tahe No. 2 was relatively uniform, and the order of proportion was main stems and leaves > lower leaves > upper leaves > middle leaves > upper stems > middle stems > lower stems.

[0122] 2023: Cotton aphids are distributed more on the stems of Tahe No. 2, and the proportion of the three parts is close to 1:1:1. The order of proportion on cotton plants is middle stems>upper stems>lower stems>main stems and leaves>upper leaves>middle leaves>lower leaves; in J8031, cotton aphids account for the largest proportion on the main stems and leaves, followed by the stems. The order of proportion on cotton plants is main stems and leaves>middle stems>upper stems>lower stems>upper leaves>middle leaves>lower leaves. However, with the passage of time, cotton aphids tend to move downward.

[0123] Overall, Xinluzhong No. 81 has a loose plant shape, II-type fruit branches, good ventilation and light transmission. When the temperature is high, cotton aphids attach to the upper part of the cotton plant. As the temperature drops, cotton aphids gradually move to the middle and upper layers. Tahe No. 2 has a cylindrical plant shape, dense fruit branches, II-type fruit branches, and no hair on the stems. The distribution of cotton aphids is relatively balanced. J8031 has a loose plant shape, II-IV-type fruit branches, more hairs, large leaves, and dark leaf color. The hairs will have a certain hindering effect, limiting the movement of cotton aphids.

[0124] 1.3 Characteristics of cotton aphid changes over time ( Figure 4 )

[0125] 2022: The survey began on the 48th day after sowing. The number of cotton aphids of different varieties showed a trend of increasing first and then decreasing during the entire survey period. The number of cotton aphids reached its peak 63 days after sowing. The number of cotton aphids of Xinluzhong No. 81 and Tahe No. 2 increased significantly, reaching 4402 and 3520 per plant, respectively. In the same period of time, the number of cotton aphids of Xinluzhong No. 81 was generally more than that of Tahe No. 2, indicating that Tahe No. 2 has better insect resistance.

[0126] 2023: The survey started on the 54th day after sowing. The number of cotton aphids of different varieties showed an increase-decrease-increase-decrease trend throughout the survey period, with two high-incidence periods of cotton aphids. The first peak of cotton aphid population was reached 67 days after sowing, with the number of cotton aphids per plant in Tahe No. 2 reaching 3288 and J8031 reaching 2344; the second peak of cotton aphid population was reached 102 days after sowing in Tahe No. 2, with the number of cotton aphids per plant reaching 2882; the second peak of cotton aphid population was reached 89 days after sowing in J8031, with the number of cotton aphids per plant reaching 1606; then the number of cotton aphids began to decline.

[0127] 1.4 Characteristics of changes in 100 cotton aphids ( Figure 5 )

[0128] Through the analysis of fixed-point survey data, it can be seen that in 2022, the number of cotton aphids among various varieties showed a trend of increasing first and then decreasing during the entire survey period. The analysis found that the number of cotton aphids continued to increase from 48 days after sowing, reaching a peak at 63 days after sowing, with the number of cotton aphids of Xinluzhong No. 81 being 303,722 per 100 plants and that of Tahe No. 2 being 80,200 per 100 plants; the rate of decrease in the number of cotton aphids slowed down after 78 days after sowing, until the number of cotton aphids gradually subsided at 88 days after sowing.

[0129] In 2023, the number of cotton aphids showed a wave-like trend among all varieties during the entire survey period, with two peak periods of cotton aphid numbers. Among them, the number of cotton aphids per 100 plants of Tahe No. 2 was more than that of J8031. The first small peak of the number of cotton aphids of the two varieties of cotton during the entire survey period was 54 days after sowing, with 54,722 aphids per 100 plants of Tahe No. 2 and 32,956 aphids per 100 plants of J8031; the second peak was reached 102 days after sowing, with 130,833 aphids per 100 plants of Tahe No. 2 and 85,667 aphids per 100 plants of J8031.

[0130] 2. The impact of cotton aphid number on cotton growth

[0131] Plant height is an important indicator for measuring cotton growth. Figure 6 It can be seen that the plant height showed a wave-like upward trend over time. In the two years, the plant height of Tahe No. 2 was significantly higher than that of Xinluzhong No. 81 and J8031. The growth of cotton at 54d and 67d after sowing in 2023 was lower than that in 2022. In 2022, the cotton grew faster in the early stage, and the growth rate of plant height slowed down at 71d after sowing; in 2023, the difference between plant heights in the early stage was significant, and the change range was large. Starting from 89d after sowing, the growth rate of plant height began to slow down until it stopped growing.

[0132] During the investigation period, the leaf area of ​​each main stem leaf showed basically the same trend over time, and the differences in leaf area of ​​main stem leaves between different leaf positions were compared. In 2022, the leaf area of ​​the 1st leaf to the 4th leaf of Xinluzhong 81 showed a "W"-shaped change over time; the 5th leaf to the 10th leaf showed a trend of first rising, then falling, and then rising again, with the largest leaf area at 63 days after sowing. The leaf area of ​​the 1st leaf to the 6th leaf of Tahe No. 2 showed a wave-like upward trend over time, with the largest leaf area at 88 days after sowing, and the 7th leaf to the 10th leaf gradually increased over time, with the largest leaf area at 78 days after sowing. In 2023, the overall change trend of the leaf area of ​​Tahe No. 2 and J8031 was the same, both showing a trend of first rising and then falling over time, and the leaf area reached the highest value at 102 days and 111 days after sowing, respectively. The leaf area of ​​each leaf position changed significantly over time, and the amplitude of change gradually decreased. The leaf area difference between leaf positions was significant, and the leaf position had a greater impact on the leaf area.

[0133] There was a significant negative correlation between the number of cotton aphids per plant and plant height, and between the number of cotton aphids per leaf and leaf area. It can be seen that the correlation between the number of cotton aphids per leaf and leaf area and plant height is stronger, and both leaf area and plant height decrease with the increase of cotton aphid number.

[0134] 3. Effect of cotton aphid quantity on cotton physiological indicators

[0135] During the investigation period, the SPAD values ​​of the main stem leaves showed basically the same trend over time. The differences in SPAD values ​​between different leaf positions were compared. In 2022, the SPAD values ​​of the 1st to 3rd leaves of Xinluzhong 81 dropped significantly over time, and the 4th to 10th leaves showed a "W" shape change, with the lowest SPAD values ​​at 63d and 88d after sowing; there were significant differences between leaf positions at 48d and 78d after sowing. The SPAD values ​​of the 1st to 4th leaves of Tahe No. 2 changed greatly, with significant differences between different times, and the lowest SPAD value at 88d after sowing; the 5th to 10th leaves showed an overall trend of decline-increase-decrease, with no significant differences between different times, and the lowest SPAD values ​​at 63d and 88d after sowing; there were significant differences in SPAD values ​​between leaf positions at 48d and 88d after sowing, and the influence of leaf position on SPAD value was more significant. In 2023, the overall change trend of Tahe No. 2 was the same, and both showed a downward-increasing-downward trend over time. The SPAD value was the highest at 67 days after sowing, and the fluctuation range was large from the 1st leaf to the 4th leaf. The SPAD value decreased significantly at 89 days after sowing, and there were significant differences among the leaf positions. The 1st leaf to the 4th leaf of J8031 showed a wave-like change over time, and the SPAD value was the lowest at 67 days and 102 days after sowing; the 5th leaf to the 10th leaf showed an "M"-shaped change over time, and the SPAD value was the highest at 67 days and 111 days after sowing; at 67 days, 79 days, and 89 days after sowing, there were significant differences in the SPAD values ​​among the leaf positions; with the aging of the leaves, the SPAD value of the lower leaf position of the cotton plant was greatly affected, and it showed an overall downward trend, and the effect on the middle and upper leaf positions was small.

[0136] During the investigation period, the change trend of single leaf water content of each main stem leaf over time was basically the same. The difference of single leaf water content between different leaf positions was compared. In 2022, the single leaf water content of Xinluzhong 81 from the 1st leaf to the 5th leaf showed an increase-decrease-increase trend over time, and was the highest at 71 days after sowing; the 6th leaf to the 10th leaf showed a "W" shape change, and the single leaf water content was the lowest at 63 days and 88 days after sowing. The single leaf water content of Tahe 2 from the 1st leaf to the 5th leaf fluctuated greatly, and showed an "M" shape change over time. The single leaf water content of each leaf position was the highest at 71 days and 88 days after sowing; the 6th leaf to the 10th leaf showed a wave-like downward trend, and the single leaf water content was the lowest at 63 days after sowing. The water content between different leaf positions was significantly different, and the leaf position had a greater impact on the single leaf water content. In 2023, the water content of single leaves of Tahe No. 2 showed an overall trend of first increasing and then decreasing over time, among which the fluctuations were larger from the 3rd leaf to the 6th leaf, and the water content of single leaves was the highest at 89 days after sowing. The trend of J8031 from the 3rd leaf to the 5th leaf was similar, showing an "M"-shaped change, and the water content of single leaves was the highest at 79 days and 102 days after sowing; from the 6th leaf to the 10th leaf, it generally increased in a wave-like manner over time. In summary, leaf position has a significant effect on leaf water content, which changes significantly over time, and the water content of each leaf position is significantly different.

[0137] There is a negative correlation between the number of cotton aphids and the SPAD value, and there is a significant negative correlation between the number of cotton aphids and the water content of single leaves. It can be seen that the number of cotton aphids has a greater impact on the water content of single leaves, and the water content of single leaves decreases with the increase of the number of cotton aphids.

[0138] The classification of the degree of cotton aphid damage to cotton is shown in Table 1. The effect of different aphid damage levels on the anthocyanin content of cotton is shown in Table 1. Figure 7 As shown in (A), the anthocyanin content in cotton leaves gradually decreased with the deepening of aphid damage, level 0>level 1>level 2>level 3. When the aphid damage level was level 0, the anthocyanin content of Tahe No. 2 was 0.76 mg / g, which was significantly higher than other levels. At level 1, it was 0.31 mg / g, which was 59.8% lower than that of level 0; at level 2, it was 0.27 mg / g, which was 64.5% lower than that of level 0; at level 3, it was 0.28 mg / g, which was 63.5% lower than that of level 0. When the damage level was level 0, the anthocyanin content of J8031 was 1.02 mg / g, and the anthocyanin content of levels 1 and 2 was 45.9% and 38.4% lower than that of level 0, respectively; there was a significant difference between it and level 3 (0.36 mg / g), which was 64.3% lower than that of level 0. In general, the anthocyanin content of J8031 was higher than that of Tahe No. 2.

[0139] Table 1

[0140] Damage level symptom Damage level Symptom Level 0 Leaves are flat and there is no cotton aphid damage Level 1 Slightly curled leaf margins Level 2 Leaf margins are curled, but less than hemispherical Level 3 The leaf margins are curled, larger than hemispherical but smaller than spherical

[0141] Effects of different aphid damage levels on cotton MDA content Figure 7 As shown in (B), the MDA content in the leaves showed a trend of first decreasing and then increasing with the deepening of the aphid damage, level 0>level 3>level 1>level 2. When the aphid damage level was level 0, the MDA content of Tahe No. 2 was 5.28μmol / g, and there was no significant difference in the MDA content in cotton leaves among the various levels; the MDA content at level 2 was 4.22μmol / g, which was 19.9% ​​lower than that at level 0. When the aphid damage level was level 0, the MDA content of J8031 was 4.19μmol / g, and the MDA content was the lowest when the aphid damage level was level 2, which was 3.32μmol / g, which was 20.6% lower than that at level 0. There was no significant difference in the MDA content in cotton leaves among the various levels. In general, the MDA content of Tahe No. 2 was higher than that of J8031.

[0142] Effects of different aphid damage levels on cotton POD activity Figure 7As shown in (C), the POD activity in cotton leaves showed a trend of first decreasing and then increasing with the deepening of aphid damage, level 3>level 2>level 1>level 0. When the aphid damage level was level 0, the POD activity of Tahe No. 2 was 253.73U / g, and there was no significant difference in POD activity between levels 2, 3 and 0, and level 1 was significantly reduced by 26.9% compared with level 0. When the aphid damage level was level 0, the POD activity of J8031 was 270.45U / g, and level 1 was significantly reduced by 10.9% compared with level 0, and there was no significant difference in POD activity between levels 2, 3 and 0. In general, the POD activity of J8031 was higher than that of Tahe No. 2.

[0143] Effects of different aphid damage levels on cotton CAT activity Figure 7 As shown in (D), the CAT activity in the leaves of Tahe No. 2 gradually increased with the deepening of aphid damage, level 3> level 2> level 1> level 0, indicating that after cotton was attacked by cotton aphids, the CAT activity in its leaves would increase to resist the damage caused by cotton aphids to the leaves; when the aphid damage level was level 0, the CAT activity was 12.15μmol / g, and when it was level 3, it was 14.98μmol / g, an increase of 23.4% compared with level 0. The CAT activity in the leaves of J8031 showed a trend of first decreasing and then increasing with the deepening of aphid damage, level 3> level 0> level 1> level 2. When the aphid damage level was level 0, the CAT activity of J8031 was 20.61μmol / g, and level 1 was significantly lower than level 0 by 11.2%. The CAT activity reached the highest level at level 3, which was 20.73μmol / g, an increase of 20.6% compared with level 0. In general, the POD activity of J8031 was higher than that of Tahe No. 2.

[0144] 4. Construct a model for monitoring the number of cotton aphids

[0145] Number of cotton aphids per plant: The monitoring model of cotton aphids was established using the Simple Linear Regression (SLR) method. The number of cotton aphids per plant of a total of 96 samples collected was used to construct the monitoring model of cotton aphids. The sample data was arranged using random numbers generated by the RAND function and divided into training set: validation set = 2:1, of which 64 were training sets and 32 were validation sets (Table 2).

[0146] Table 2

[0147]

[0148] Number of cotton aphids per 100 plants: The monitoring model of cotton aphids was established using the Simple Linear Regression (SLR) method. The number of cotton aphids per 100 plants of the 82 samples collected was used to construct the monitoring model of cotton aphids. The sample data was arranged using random numbers generated by the RAND function and divided into training set: validation set = 2:1, of which 54 were training sets and 28 were validation sets (Table 3).

[0149] Table 3

[0150]

[0151] The correlation analysis between the number of cotton aphids on the main stem leaves and the number of cotton aphids per plant (E0) was performed. Figure 8 As shown in (A), the results show that the number of cotton aphids in each leaf position is positively correlated with the number of cotton aphids per plant; among them, the fourth leaf, the sixth leaf, and the seventh leaf are highly correlated with the number of cotton aphids, and the correlation coefficients are all greater than 0.70. The correlation analysis of the number of cotton aphids in different parts and the number of cotton aphids per plant was carried out separately, as shown in Figure 2. Figure 8 As shown in (B), the results show that the number of cotton aphids in each part is positively correlated with the number of cotton aphids per plant; L3, S2 and S3 are less correlated with the number of cotton aphids per plant, among which L0, L1, L2 and S1 are more correlated with the number of cotton aphids per plant, and the correlation coefficients are all greater than 0.70, and the correlation L0=S1>L1>L2>L3>S2>S3; This shows that the number of cotton aphids in the main stem leaves, upper stems, upper leaves and other parts can reflect the number of cotton aphids per plant, and thus the rapid acquisition of cotton aphid occurrence information can be achieved. The correlation analysis between the number of cotton aphids in different parts and the number of cotton aphids per 100 plants was carried out separately, as shown in Figure 8 As shown in (C), the results show that the number of cotton aphids in each part is positively correlated with the number of cotton aphids per 100 plants, and the correlation between leaves and the number of cotton aphids per 100 plants is relatively good; the number of cotton aphids in different parts can better reflect the number of cotton aphids per 100 plants, and thus the rapid acquisition of cotton aphid occurrence information can be achieved.

[0152] Model of the number of cotton aphids per plant: Through correlation analysis, four variables with high correlation with the number of cotton aphids per plant were selected to construct a univariate linear regression model (SLR), as shown in Table 4. In the SLR model that uses a single variable to monitor the number of cotton aphids per plant, the model has the highest accuracy when the training object is S1, R 2 The model was verified, among which: the model constructed with S1 had the best accuracy, and its R 2 =0.92, RMSE=566.18 heads / plant, RPD=1.76, the measured values ​​and the predicted values ​​showed a good fitting relationship. In summary, the model constructed with S1 as the variable had the best effect.

[0153] Table 4

[0154] Model of the number of cotton aphids per 100 plants: Through correlation analysis, four variables with high correlation with the number of cotton aphids per 100 plants were selected to construct a univariate linear regression model (SLR) as shown in Table 5. In the SLR model that uses a single variable to monitor the number of cotton aphids per 100 plants, the model has high accuracy when the training object is L2, and R 2 The model was verified by L2, and the accuracy of the model was better. 2 =0.76, RMSE=25 227.48 heads / 100 plants, RPD=1.78, and the measured values ​​and predicted values ​​showed a good fitting relationship. In summary, the model constructed with L2 as the variable had the best effect.

[0155] Table 5

[0156] Fig. 9 is the reflectance of cotton leaves under different numbers of cotton aphids. In the range of 757-1000nm, the leaf reflectance is the highest when the number of cotton aphids is 219. The leaf reflectance shows a trend of first decreasing, then increasing, and then decreasing as the number of cotton aphids increases.

[0157] The wavelength range of the hyperspectral spectrum obtained in the experiment is 320-1100nm (ultraviolet-visible-near infrared), with a total of 732 samples. During the growth process of cotton, the leaves and physiological growth characteristics will show regular changes, which will affect the spectral reflectance of the leaves. Fig.10It can be seen that after the original spectrum is processed by convolution smoothing, its spectral curve is basically consistent with the change trend of the original spectral curve. The convolution smoothing process further smoothes and filters the hyperspectral curve. Through this process, the details of the hyperspectral curve are clearer, the signal-to-noise ratio is optimized, and the data is more readable and interpretable. The spectral curve after the first-order differential transformation has extreme points near 520, 694, 713, 719, 761, and 933nm, and reaches a maximum at 716nm. This may be due to the different numbers of cotton aphids causing different degrees of damage to the leaves, resulting in differences in the reflected spectral curves. In the range of 750-800nm, the first-order differential spectrum value is close to 0, indicating that the spectral reflectance caused by different numbers of cotton aphids reaches an extreme value, and peaks appear in the green light (500-560nm) and red light (670-800nm) bands. The value reflects the reflectivity of different numbers of cotton aphids to cotton leaf chlorophyll under the corresponding light wave. The spectral data after second-order differential processing fluctuates frequently in the range of 680-780nm, especially in the range of 899-1000nm, which indicates that the spectrum is in the absorption zone in this range. Overall, compared with the first-order differential, many peaks and troughs appear after the second-order differential transformation of the spectral data. Compared with the original spectrum, the spectral curve after reciprocal transformation has obvious peak-valley inversion phenomenon. The blue valley and red valley become blue peaks and red peaks, the green peak becomes green valley, the reflection peak becomes the absorption valley, and the reflectivity of the high reflection platform becomes a smooth curve close to 0. In the visible light range of 400-700nm, the fluctuation amplitude of the spectral curve becomes larger, the concavity of the peaks and troughs deepens, and the difference between the spectral curves increases significantly. Compared with the original spectrum, the spectral values ​​after different spectral transformation forms are different in some subtle bands, which effectively highlights and amplifies some noise that is not obvious in the original spectrum.

[0158] The correlation between the single leaf spectral data after spectral transformation and the number of cotton aphids is shown in the figure below: Fig.11 As shown. In the original spectrum at 400-724nm, the number of cotton aphids is negatively correlated with the reflectance of a single leaf, and the curve fluctuates greatly; while in 724-1000nm, it is positively correlated, the curve fluctuates less, and shows a gradually rising trend. The spectral curve processed by the first-order differential and the second-order differential changes in a wavy shape, alternating between positive and negative values, with a correlation coefficient between -0.3 and 0.3, and the curve fluctuates greatly. The reciprocal spectral value is positively correlated with the number of cotton aphids and the reflectance of a single leaf at 400-724nm, and negatively correlated at 724-1000nm.

[0159] The total sample number, training set, and validation set for determining the number of cotton aphids on a single leaf in this embodiment are shown in Table 6.

[0160] Table 6

[0161] The screening results of the characteristic bands of the single leaf spectrum are shown in Table 7. The sensitive bands of the single leaf spectrum are mostly distributed in the near-infrared region, with a small amount distributed in the visible light region. The characteristic bands screened by Pearson after RS, SG, and LG processing are concentrated in the range of 956-967nm, and most of them are adjacent bands; while the characteristic bands screened by Pearson after FD, SD, and RG processing are concentrated in the range of 522-808nm. In summary, different dimensionality reduction methods can reduce the dimensionality of data after different preprocessing to a certain extent.

[0162] Table 7

[0163]

[0164]

[0165] Fig.12 The stratified reflectance of cotton under different numbers of cotton aphids. In the range of 700-1000nm, the stratified spectral reflectance gradually decreases with the decrease of the cotton stratification position, and the reflectance is upper layer>middle layer>lower layer. The number of cotton aphids also gradually decreases with the decrease of sampling position. At 680-750nm, the red edge position shows a "blue shift" phenomenon as the aphid damage deepens, and the spectral reflectance between different layers is significantly different.

[0166] The spectral characteristics of different pre-processing layers are as follows: Fig.13As shown in the figure, the original hyperspectral feature has an absorption valley at 492nm and an absorption valley at 670nm. There is a green light band between the two valleys. Due to the low absorption of light, a reflection peak is formed at 552nm. The overall curve shows a "root" trend. In the range of 700-750nm, the reflectivity increases sharply and then changes steadily, forming a high reflection platform. After the original spectrum is processed by convolution smoothing, its spectral curve is basically consistent with the change trend of the original spectral curve. The convolution smoothing process further smoothes and filters the hyperspectral curve, thereby reducing noise and unnecessary fluctuations. Through this process, the details of the hyperspectral curve are clearer, and the ratio of signal to noise is optimized, making the data more readable and interpretable. The spectral curve after the first-order differential transformation has extreme points near 527, 566, 719, and 930nm, and reaches a maximum at 719nm. The curve fluctuations in the range of 669-974nm are relatively violent. This phenomenon may be due to the different degrees of damage caused by different numbers of cotton aphids on the leaves, and the different light conditions at different layer heights lead to differences in the reflected spectral curves. In the range of 801-917nm, the first-order differential spectrum value is close to 0, indicating that the spectral reflectance caused by different numbers of cotton aphids reaches an extreme value, and peaks appear in the green light (500-560nm) and red light (670-800nm) bands. The peaks are very significant, and the differences between different numbers are obvious. The spectral data after second-order differential processing fluctuates frequently in the range of 781-1 000nm, especially in the range of 923-1000nm. The change range is more significant, indicating that the spectrum is in the absorption zone in this range. Overall, compared with the original spectrum, the spectrum curve after the inverse transformation has obvious peak-valley inversion phenomenon. The blue valley and red valley become blue peak and red peak, the green peak becomes green valley, the reflection peak becomes absorption valley, and the reflectivity of the high reflection platform becomes a smooth curve close to 0. In the visible light range of 400-700nm, the fluctuation amplitude of the spectrum curve becomes larger, the concavity of the peak and trough deepens, and the difference between the spectrum curves increases significantly. Overall, compared with the original spectrum, the spectrum values ​​after different spectral transformation forms effectively highlight and amplify the feature points that are not obvious in the original spectrum.

[0167] The correlation between the stratified spectral data after spectral transformation and the number of cotton aphids is shown in the figure below: Fig.14As shown. The original spectrum and the spectrum after convolution smoothing are positively correlated with the number of cotton aphids and the stratified reflectance in the whole band. The curve fluctuates greatly and the correlation coefficient is between 0.03 and 0.26. The spectral curve processed by the first-order differential and the second-order differential changes in a wavy shape, alternating between positive and negative values, with a correlation coefficient between -0.3 and 0.3. The curve fluctuates greatly and is closely correlated with the number of cotton aphids. The spectral value after the second-order transformation is more closely correlated with it. The reciprocal spectral value is positively correlated with the number of cotton aphids in the range of 400-723nm and negatively correlated with the number of cotton aphids in the range of 724-1000nm. The correlation coefficient of the whole band varies from -0.13 to 0.11. The spectral value after logarithmic transformation and the spectral value after reciprocal transformation show completely opposite trends.

[0168] The total sample number, training set, and validation set for determining the number of cotton aphids in each layer in this example are shown in Table 8.

[0169] Table 8

[0170] The results of the screening of the characteristic bands of the stratified spectrum are shown in Table 9. The sensitive bands of the stratified spectrum are mostly distributed in the near-infrared region, with a small amount distributed in the visible light region. The characteristic bands screened by Pearson after RS, SG, LG, and RG processing are concentrated in the range of 956-1000nm, and most of them are adjacent bands; while the characteristic bands screened by Pearson after FD and SD processing are concentrated in the visible light range. In general, different dimensionality reduction methods can reduce the dimensionality of data after different preprocessing to a certain extent.

[0171] Table 9

[0172]

[0173]

[0174] Single-leaf cotton aphid population monitoring model:

[0175] (1) The characteristic bands extracted by different band screening methods after different preprocessing are used as model input values ​​to establish the LR model. As shown in Table 10, the R 2 From 0.03 to 0.28, RMSE decreased from 109.17 heads / leaf to 94.04 heads / leaf, and RPD increased from 0.18 to 0.63. Modeling effect LG+SPA>SG+SPA>

[0176] RG+SPA>FD+Pearson>RS+SPA>SD+Pearson>RG+Pearson>SG+Pearson>FD+SPA>SD+SPA>RS+Pearson>LG+Pearson. Among them, the model constructed by LG+SPA has the best effect, but there is a certain difference between the accuracy of its validation set and the accuracy of the training set. Fig.15 It can be seen that the verification effect of FD+Pearson model is better, R 2 The FD+Pearson model has the best effect in the estimation model of cotton aphid population based on LR.

[0177] Table 10

[0178]

[0179] (2) The XGBoost model was established using the feature bands extracted by different band screening methods after different preprocessing as the model input values. As shown in Table 11, the XGBoost model under different treatments has better effect than RS after FD and SD preprocessing. The model constructed by Pearson method has better effect than RS. 2 The values ​​were 0.79 and 0.81, the RMSE values ​​were 53.50 heads / leaf and 53.35 heads / leaf, and the RPD values ​​were 1.43 and 1.42. Compared with RS, the model constructed by SPA method had better effect after SG and LG preprocessing. 2 The results of model validation are shown in Figure 2. The results of model validation are shown in Figure 2. The results of model validation are shown in Figure 2. The results of model validation are shown in Figure 2. The results of model validation are shown in Figure 2. The results of model validation are shown in Figure 2. The results of model validation are shown in Figure 2. Fig.16 , the FD+Pearson model has higher accuracy, and its R 2 The RMSE is 41.71 head / leaf, and the RPD is 1.48, followed by SD+Pearson. 2 The value of the training set is 0.66, the RMSE is 43.96 heads / leaf, and the RPD is 1.41. The results of the model training set and the validation set are consistent.

[0180] Table 11

[0181]

[0182] (3) The characteristic bands extracted by different band screening methods after different pretreatments were used as model input values ​​to establish the PLSR model. As shown in Table 12, the R2 From 0.03 to 0.28, RMSE decreased from 109.16 heads / leaf to 94.27 heads / leaf, and RPD increased from 0.18 to 0.62. Modeling effect LG+SPA>RG+SPA>

[0183] FD+Pearson>SG+SPA>RS+SPA>SD+Pearson>RG+Pearson>FD+SPA>SD+SPA>

[0184] SG+Pearson>RS+Pearson>LG+Pearson. The model validation results are as follows Fig.17 , the FD+Pearson model has higher accuracy, and its R 2 The RPD of each model is less than 1.4, and the prediction accuracy of the model is low.

[0185] Table 12

[0186]

[0187]

[0188] Layered cotton aphid population monitoring model:

[0189] (1) The characteristic bands extracted by different band screening methods after different preprocessing were used as model input values ​​to establish the LR model. As shown in Table 13, the LR models under different treatments have better effects than RS after FD preprocessing. The model constructed by Pearson method has better R 2 The modeling effect is RS+SPA>RG+SPA>LG+SPA>FD+Pearson>SG+SPA>RS+Pearson>SD+SPA>

[0190] FD+SPA>RG+Pearson>SD+Pearson>LG+Pearson>SG+Pearson. Depend on Fig.18 It can be seen that the RS+SPA model has the best verification effect. 2 The RR is 0.75, the RMSE is 226.73 heads / layer, and the RPD is 1.65. The results of the model training set and the validation set are consistent. In addition, the RG+SPA, LG+SPA, and FD+Pearson models are all highly accurate and can be used to predict the occurrence of cotton aphids. In summary, among the cotton aphid population estimation models based on LR, the RS+SPA model has the best effect.

[0191] Table 13

[0192]

[0193] (2) The XGBoost model was established by using the feature bands extracted by different band screening methods after different preprocessing as the model input values. As shown in Table 14, the XGBoost model under different treatments has better effect than RS after SG, FD, and SD preprocessing. The model constructed by Pearson method has better effect than RS. 2 The results are 0.91, 0.96, and 0.95, respectively. The RMSE are 265.90 heads / leaf, 132.62 heads / leaf, and 157.78 heads / leaf, respectively. The RPD are 1.44, 3.99, and 3.25, respectively. Compared with RS, the model constructed by SPA method has better effect after LG and RG preprocessing. 2 The results are 0.97 and 0.98 respectively, RMSE are 141.49 heads / leaf and 128.41 heads / leaf respectively, and RPD are 3.51 and 3.96 respectively. The modeling effect is RG+SPA>FD+Pearson>LG+SPA>SD+Pearson>RS+SPA>SG+SPA>SG+Pearson>RG+Pearson>RS+Pearson>LG+Pearson>SD+SPA>FD+SPA. Fig.19 It can be seen that the verification effect of RG+SPA model is the best. 2 The model training set and validation set results were consistent. In the cotton aphid population estimation model based on SVR, the modeling accuracy of each combination was good and could be used to predict the occurrence of cotton aphid population. In summary, the RG+SPA model performed best in the cotton aphid population estimation model based on SVR.

[0194] Table 14

[0195]

[0196] (3) The characteristic bands extracted by different band screening methods after different pretreatments were used as model input values ​​to establish the PLSR model. As shown in Table 15, the PLSR models under different treatments have better effects than RS after FD pretreatment. The model constructed by Pearson method has better effects than RS. 2 The modeling effect is RS+SPA>LG+SPA>FD+Pearson>RG+SPA>SG+SPA>FD+SPA>SD+SPA>

[0197] SD+Pearson>LG+Pearson>RG+Pearson>RS+Pearson>SG+Pearson. Depend on Fig. 20 It can be seen that the RS+SPA model has the best verification effect. 2 The RS+SPA model has a high accuracy and can be used to predict the occurrence of cotton aphids. In summary, among the cotton aphid estimation models based on PLSR, the RS+SPA model has the best effect.

[0198] Table 15

[0199]

[0200]

[0201] Although the embodiments of the present invention have been disclosed as above, they are not limited to the applications listed in the specification and the implementation modes, and they can be fully applied to various fields suitable for the present invention. For those familiar with the art, additional modifications can be easily implemented. Therefore, without departing from the general concept defined by the claims and the scope of equivalents, the present invention is not limited to the specific details and the illustrations shown and described herein.

Claims

1. A method for evaluating cotton aphid situation based on hyperspectral technology, characterized in that: The following steps are involved: Step 1: Divide the cotton test site into a destructive sampling area and a non-destructive sampling area for sampling; Step 2: Determine the number data of cotton aphids, the hyperspectral data of cotton after aphid infestation, the growth index of cotton, and the physiological index data; Step 3: Analyze the data obtained in step 2 to obtain the temporal and spatial distribution of the number of cotton aphids, the change characteristics of the hyperspectral reflectance of cotton after aphid infestation, and the change characteristics of cotton growth indicators and physiological indicators; Step 4: Process the hyperspectral data of cotton after aphid infestation obtained in step 2, screen the spectral sensitive bands, and build and verify the cotton aphid population monitoring model.

2. The method for evaluating cotton aphid situation based on hyperspectral technology according to claim 1, characterized in that: In the step 1, the sampling method of the destructive sampling area is random sampling, and each time sampling is performed, 5 cotton plants with uniform growth are randomly selected from each variety in this area; the sampling method of the non-destructive sampling area is to randomly select 4 to 6 fixed points for each variety in this area, and each fixed point selects 3 to 5 consecutive cotton plants with uniform growth as continuous survey points, and marks them with signs.

3. The method for evaluating cotton aphid situation based on hyperspectral technology according to claim 2, characterized in that: In the step 2, the specific method for determining the number of cotton aphids is: S21. Destructive sampling area: The cotton plants were divided into three parts: upper layer, middle layer and lower layer according to plant height; the main stem leaves of each plant were removed, the first main stem leaf at the bottom was the inverted leaf, and the remaining main stem leaves were marked from bottom to top; the total number of cotton aphids on each main stem leaf, as well as the total number of cotton aphids on the upper leaves, middle leaves, lower leaves and upper stems, middle stems and lower stems were counted; S22. Non-destructive sampling area: Investigate the total number of cotton aphids on each cotton plant at each designated point and convert it into the number of aphids per 100 plants.

4. The method for evaluating cotton aphid situation based on hyperspectral technology according to claim 2, characterized in that: In the step 2, the specific method for determining the hyperspectral data of cotton after aphid infestation is: using a PSR-1100f portable ground spectrometer to collect spectra, the spectral measurement range is 320-1100nm, the probe has a field of view of 25°, the resolution is 3.2nm, the canopy spectrum is scanned more than 10 times on average, and white board correction is required every 10-15 minutes during spectral data collection. The white board is placed flat, and the surface of the white board cannot be blocked by shadows during correction; Destructive sampling area: collect the spectral reflectance of each main stem leaf, avoid the leaf veins when collecting data, divide each leaf into three parts: upper, left and right, collect one spectral curve for each part, and perform whiteboard calibration before each spectrum collection and measurement; Non-destructive sampling area: collect canopy spectra of cotton plants at each fixed point; All spectral measurements were performed from a height of 50 cm above the cotton plants, and spectral data were acquired under clear skies with minimal or no wind between 11:00 and 14:

00.

5. The method for evaluating cotton aphid situation based on hyperspectral technology according to claim 2, characterized in that: In the step 2, the cotton growth indicators include: plant height and leaf area; the cotton physiological indicators include: chlorophyll relative content value, single leaf moisture content, anthocyanin content, malondialdehyde content, peroxidase activity and catalase activity.

6. The method for evaluating cotton aphid situation based on hyperspectral technology according to claim 2, characterized in that: In step 4, the specific method for processing the hyperspectral data of cotton after aphid infestation is: S41, hyperspectral data preprocessing: spectral data preprocessing using convolution smoothing; S42. Spectral data transformation: The spectral data is processed by using first-order differential, second-order differential, logarithmic and inverse transformations.

7. The method for evaluating cotton aphid situation based on hyperspectral technology according to claim 2, characterized in that: In the step 4, the specific method for screening the spectral sensitive bands is: using the Pearson correlation coefficient method and the continuous projection method to screen the characteristic bands related to the number of cotton aphids.

8. The method for evaluating cotton aphid situation based on hyperspectral technology according to claim 2, characterized in that: In step 4, the specific method for constructing and verifying the cotton aphid quantity monitoring model is: Construct a cotton aphid population monitoring model: linear regression, extreme gradient boosting tree and partial least squares regression were used to construct a single leaf cotton aphid population monitoring model and a stratified cotton aphid population monitoring model; Sample set division and model accuracy verification: The RAND function was used to generate random numbers, and the collected samples were randomly sorted and divided into training set: verification set = 2:

1. Three indicators, namely, determination coefficient, root mean square error and relative analytical error, were selected to analyze the model accuracy.

9. The method for evaluating cotton aphid situation based on hyperspectral technology according to claim 6, characterized in that: The calculation formula of the first-order differential is: The calculation formula of the second-order differential is: In the formula, R' λ(i) is the first-order differential value at wavelength i, R λ(i+1) and R λ(i-1) are the spectral reflectance values ​​at wavelengths i+1 and i-1, respectively, R” λ(i) is the second-order differential at wavelength i, R' λ(i+1) and R' λ(i-1) are the first-order differential values ​​at wavelengths i+1 and i-1 respectively, and 2Δλ represents the differential interval.

10. The method for evaluating cotton aphid situation based on hyperspectral technology according to claim 6, characterized in that: The calculation formula of the logarithm is: R(i)=ln(r(i)); In the formula, i represents the number of bands, r(i) represents the i-th spectral reflectance, and R(i) represents the i-th spectral reflectance after logarithmic transformation, where i = 400-1000; The calculation formula of the reciprocal is: Wherein, i represents the number of bands, r(i) represents the i-th spectral reflectance, and R(i) represents the i-th spectral reflectance after reciprocal transformation, where i=400~1000.

Images