A hyperspectral identification method for cotton pesticide damage and spider mite pests

Abstract

The present invention belongs to the field of agricultural technology, and specifically relates to a hyperspectral identification method for cotton pesticide damage and spider mite pests. The method is based on ground spectral data, takes cotton plants with pesticide damage and spider mite pests as research objects, uses a physical spectrometer to extract cotton canopy spectral data from the cotton seedling stage to the flowering and boll stage, analyzes the differences between the cotton pesticide damage spectrum and the spider mite pest spectrum; and uses unmanned aerial vehicle hyperspectral remote sensing image data to accurately and quickly analyze cotton fields; and then constructs a cotton hyperspectral pesticide damage and insect pest hyperspectral image identification model to achieve accurate discrimination of cotton pesticide damage and spider mite pests at the field scale and distribution area statistics, which is helpful to grasp the occurrence of insect pests and the use of drugs in cotton fields, and then guides the field plant protection management of cotton from the seedling stage to the flowering and boll stage, achieves scientific reduction of drugs, accurate use of drugs and disinfection, saves drug costs, and improves the quality of the ecological environment.

Description

Technical Field

[0001] The invention belongs to the technical field of agriculture, and in particular relates to a high-spectrum identification method for cotton pesticide damage and spider mite pests. Background Art

[0002] Cotton is a major cash crop in my country. Xinjiang's cotton planting area and output have ranked first in the country for 27 consecutive years. It is a high-quality cotton producing area in my country. The cotton industry has played an important role in achieving desert greening, increasing farmers' income, and developing the national economy. Due to the characteristics of upland cotton that are easy to harvest mechanically, Xinjiang has gradually vigorously promoted the planting of upland cotton (short-staple cotton) during the development of agricultural mechanization. Compared with sea island cotton (long-staple cotton), upland cotton is more prone to diseases and insect pests during its growth. There are more than 10 kinds of pests that cause harm all year round. Insect pests cause cotton production to reduce by about 10-15%. In addition, due to the influence of global climate and environmental changes and the weakening of natural ecological regulation of cotton fields due to continuous planting, cotton insect pests have a tendency to become increasingly severe.

[0003] At present, the prediction and forecast of cotton diseases and pests mainly rely on the field survey experience and judgment of technical personnel from local agricultural plant protection departments, which is time-consuming and labor-intensive. At the same time, due to the limited number and area of ​​survey plots, it is difficult to accurately reflect the actual occurrence and damage dynamics in the field. The current prevention and control mode of diseases and pests is generally passive large-scale spraying of pesticides, which cannot achieve timely and rapid precise prevention and control of insect sources, and increases the application range and dosage of pesticides. Excessive chemical pesticides will cause negative effects such as mutagenicity, carcinogenicity, teratogenicity, pesticide residues and environmental pollution.

[0004] After searching, the inventor found that the prior art still has the following technical problems:

[0005] 1) Currently, the monitoring of crop diseases and pests mostly uses satellite remote sensing images and drone remote sensing images. Research on cotton diseases and pests is mostly focused on spider mite pests, while the impact of cotton pesticide damage on the cotton growth process is ignored, and no investigation has been conducted on the area and extent of cotton pesticide damage.

[0006] 2) The areas affected by spider mites and pesticide damage are confused based on image features, resulting in misjudgment, which will increase the use of pesticides and further aggravate the level of cotton pesticide damage, which is not conducive to the precise prevention and control of cotton pests and has a negative impact on cotton yields. Summary of the invention

[0007] In order to solve the defects of the above-mentioned prior art, the purpose of the present invention is to provide a high-spectral identification method for cotton pesticide damage and spider mite pests, which can realize the accurate identification of cotton pesticide damage and spider mite pests at the field scale and the statistics of distribution areas, which is helpful to grasp the occurrence of pests and the use of pesticides in cotton fields, and then guide the field plant protection management of cotton from the seedling stage to the flowering and boll stage, realize scientific reduction of pesticides, precise use of pesticides for disinfection, save pesticide costs, and improve the quality of the ecological environment.

[0008] The technical solution adopted by the present invention is as follows:

[0009] A high-spectral identification method for cotton pesticide damage and spider mite pests comprises the following steps:

[0010] Step 1: Obtain cotton canopy spectral data from seedling stage to flowering and boll stage through ground object spectrometer and clean it, analyze the spectral characteristics of cotton pesticide damage and spider mite pests, and screen out cotton canopy spectral data of pesticide damage and spider mite pests;

[0011] Step 2: Use a low-altitude UAV equipped with an imaging hyperspectral instrument to obtain hyperspectral images of the cotton canopy during the flowering and boll period. After data processing, complete the index calculation;

[0012] Step 3: Construct a hyperspectral recognition model for pesticide damage and spider mite pests based on spectral features, apply the model to UAV hyperspectral images, perform feature determination, obtain the distribution map of pesticide damage and spider mite pests at the field scale, and output the final results.

[0013] The above-mentioned technical scheme is adopted, based on ground spectral data, and cotton plants with pesticide damage and spider mite pests are taken as the research objects. The physical spectrometer is used to extract the cotton canopy spectral data from the cotton seedling stage to the flowering and boll stage, and the differences between the cotton pesticide damage spectrum and the spider mite pest spectrum are analyzed; and the unmanned aerial vehicle hyperspectral remote sensing image data are used to accurately and quickly analyze the cotton field plots; and then a cotton hyperspectral pesticide damage and insect pest hyperspectral image identification model is constructed, which can realize the accurate identification of cotton pesticide damage and spider mite pests at the field scale and the distribution area statistics, which is helpful to grasp the occurrence of insect pests and the use of drugs in cotton fields, and then guide the field plant protection management of cotton from the seedling stage to the flowering and boll stage, realize scientific reduction of pesticides, precise use of pesticides and disinfection, save pesticide costs, and improve the quality of the ecological environment.

[0014] Furthermore, in step 2, the UAV obtains the cotton canopy hyperspectral data set of the test area, and after processing with SpecView, Pix4D, ENVI, and PIE software, a hyperspectral image with a ground resolution of 1 cm is obtained; after Savitzky-Golay smoothing and spectral differential processing of the image, standard preprocessed data is obtained, and the ENVI ROI tool is used for cropping to obtain the hyperspectral image data of the target plot.

[0015] Furthermore, the step 3 specifically includes:

[0016] Step 3.1: After the spectral data and hyperspectral images are processed by the PCA method, the correlation between each factor is analyzed one by one, and the hyperspectral recognition model for pesticide damage and spider mite pests is constructed by fitting the partial least squares regression analysis model. The calculation formula is as follows:

[0017] X=TP T +E;

[0018] Y=UQ T +F;

[0019] Among them, X is the n×m prediction matrix; Y is the n×p response matrix; T is the n×l X projection matrix; U is the n×l Y projection matrix; P is the m×l orthogonal loading matrix; Q is the p×l orthogonal loading matrix; matrices E and F are error terms, which are normal random variables that obey independent and identical distributions;

[0020] Step 3.2: Perform data assimilation using the 4DVar cost function, taking into account the changes in observations over time scales. The assimilation formula is as follows:

[0021]

[0022] Among them, x k represents the model parameters, B is the model error, y is the number of observations, Q is the observation error, and H is the observation operator;

[0023] Step 3.3: Assign the data to the same spatial scale and apply it to the processed hyperspectral image data to identify whether the factor of cotton stress is pesticide damage or spider mite pests, and obtain the cotton pesticide damage distribution data and spider mite pest distribution data at the field scale.

[0024] Furthermore, the method for constructing the hyperspectral recognition model for pesticide damage and spider mite pests in step 3 includes:

[0025] 1) Obtain ground hyperspectral data;

[0026] 2) Use Savitzky-Golay to clean and denoise the data;

[0027] 3) Perform derivative and differential calculations. The derivative is calculated using

[0028] and Formula model for calculation;

[0029] Among them, R 1st (λ i ) is the spectral band λ i The first-order derivative value on R(λi ) is the reflectivity value at band i, R(λ i+1 ) is the reflectivity value at the (i+1) band; R 2nd (λ i ) is the spectral band λ i The second-order derivative value on R(λ i ) is the reflectivity value at band i, R(λ i+1 ) is the reflectivity value at the (i+1) band;

[0030] Differentiation and ρ″(λ i )=[ρ′(λ i+1 )-ρ′(λ i-1 )] / 2(λ i+1 -λ i-1 ) formula model for calculation;

[0031] Among them, ρ′(λ i ) is the spectral band λ i The first-order differential value on ρ(λ i+1 ) is the reflectivity value at the (i+1) band; ρ(λ i-1 ) is the reflectivity value at the (i-1) band; ρ″(λ i ) is the spectral band λ i The second-order differential value on ρ′(λ i+1 ) is the first-order differential value at the (i+1) band; ρ′(λ i-1 ) is the first-order differential value at the (i-1) band; λ i is the wavelength value of band i; λ (i+1) is the wavelength value of the (i+1) band.

[0032] 4) Characteristic spectrum analysis: analyze the cotton canopy spectral data, derivative and differential data, screen sensitive spectrum segments, and extract characteristic spectrum segments;

[0033] 5) Based on the extracted characteristic spectra of cotton pesticide damage and spider mite pests, the partial least squares model was used to construct the recognition model.

[0034] Furthermore, the ground object spectrometer is a SVC HR768 portable ground object spectrometer with a spectrum range of 350nm-2500nm, 768 band channels, and a spectral resolution of 3.5nm@1000nm.

[0035] Furthermore, the data collection requirements in step 1 are:

[0036] 1) The spectral data of each point of the sampling data shall be no less than 10 times;

[0037] 2) The ground visibility is not less than 10km; the amount of light cumulus clouds within a 90° solid angle around the sun should be less than 2%, with no cirrus clouds or cumulus congestus clouds; the wind force should be less than level 3;

[0038] 3) The portable ground object spectrometer uses a 25° field of view optical fiber probe to collect ground object spectral data.

[0039] In summary, due to the adoption of the above technical solution, the beneficial effects of the present invention are:

[0040] Based on ground spectral data, cotton plants with pesticide damage and spider mite pests were taken as the research objects, and the physical spectrometer was used to extract the cotton canopy spectral data from the seedling stage to the flowering and boll stage, and the differences between the cotton pesticide damage spectrum and the spider mite pest spectrum were analyzed; and the unmanned aerial vehicle hyperspectral remote sensing image data were used to accurately and quickly analyze the cotton fields; and then a cotton hyperspectral pesticide damage and insect pest hyperspectral image identification model was constructed to achieve accurate identification of cotton pesticide damage and spider mite pests at the field scale and distribution area statistics, which is helpful to understand the occurrence of insect pests and the use of pesticides in cotton fields, and then guide the field plant protection management of cotton from the seedling stage to the flowering and boll stage, achieve scientific reduction of pesticides, precise use of pesticides and disinfection, save pesticide costs, and improve the quality of the ecological environment. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] The present invention will now be described by way of example with reference to the accompanying drawings, in which:

[0042] Figure 1 : is a technical flow chart of a high-spectrum identification method for cotton pesticide damage and spider mite pests in an embodiment of the present invention;

[0043] Figure 2 It is a flow chart of a method for constructing a characteristic spectrum determination model for cotton pesticide damage and spider mite pests in an embodiment of the present invention;

[0044] Figure 3 This is a flow chart of a method for determining cotton pesticide damage and spider mite pest areas using drone hyperspectral images in an embodiment of the present invention;

[0045] Figure 4 It is a formula list diagram in an embodiment of the present invention;

[0046] Figure 5 is a canopy spectrum curve diagram of cotton plants under stress of different pesticide damage levels in an embodiment of the present invention;

[0047] Figure 6 It is a canopy spectrum curve diagram of cotton plants under stress of different spider mite pest levels in the embodiment of the present invention. DETAILED DESCRIPTION

[0048] To make the purpose, technical scheme and advantages of the embodiments of the present application clearer, the technical scheme in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all of the embodiments. The components of the embodiments of the present application usually described and shown in the drawings here can be arranged and designed in various configurations. Therefore, the following detailed description of the embodiments of the present application provided in the drawings is not intended to limit the scope of the application claimed for protection, but merely represents the selected embodiments of the present application. Based on the embodiments of the present application, all other embodiments obtained by those skilled in the art without making creative work belong to the scope of protection of the present application.

[0049] Combine the following Figure 1 to Figure 6 The present invention is described in detail.

[0050] Refer to the attached Figure 1 , Attachment Figure 2 , Attachment Figure 3 The present invention provides a hyperspectral identification method for cotton pesticide damage and spider mite pests, performs remote sensing monitoring and ground field surveys on cotton planting areas from the seedling stage to the flowering and boll stage in May, and then obtains the distribution results of pesticide damage and spider mite pests at the field scale. The specific implementation steps are as follows:

[0051] S22. Use a portable ground spectrometer to obtain cotton canopy spectral data from the seedling stage to the flowering and boll stage;

[0052] S32, cleaning the cotton canopy spectral data and removing the abnormalities;

[0053] S42, analyzing the differences in canopy spectral characteristics of cotton pesticide damage and spider mite pests, and screening out the canopy sensitive spectral segments of cotton pesticide damage and spider mite pests;

[0054] S52. Construct a hyperspectral recognition model for pesticide damage and spider mite pests based on spectral features;

[0055] S21. Obtain hyperspectral images of cotton canopy from seedling stage to flowering and boll stage by using a low-altitude UAV equipped with a hyperspectral imager;

[0056] S31, performing data processing and geo-referencing on the cotton canopy hyperspectral image;

[0057] S41, performing index calculation;

[0058] S51. Make a preliminary assessment of the areas where pesticide and insect pests occur;

[0059] S6. Apply the model to UAV hyperspectral images to determine the characteristics of areas where pesticide damage and insect pests occur;

[0060] S7 and S8, obtain the distribution map of pesticide damage and spider mite pests at the field scale;

[0061] S9. Output the final result.

[0062] In this embodiment, in step S21, the drone obtains the cotton canopy hyperspectral dataset of the test area, and after processing by Spec View, Pix4D, ENVI, and PIE software, a hyperspectral image with a ground resolution of 1 cm is obtained; after Savitzky-Golay smoothing and spectral differential processing are performed on the image, standard preprocessed data is obtained, and the ENVI ROI tool is used for cropping to obtain the hyperspectral image data of the target plot.

[0063] In this embodiment, the step S52 specifically includes:

[0064] Step 3.1: After the spectral data and hyperspectral images are processed by the PCA method, the correlation between each factor is analyzed one by one, and the hyperspectral recognition model for pesticide damage and spider mite pests is constructed by fitting the partial least squares regression analysis model. The calculation formula is as follows:

[0065] X=TP T +E;

[0066] Y=UQ T +F;

[0067] Among them, X is the n×m prediction matrix; T is the n×l X projection matrix; P is the m×l orthogonal load matrix; matrix E is the error term, which obeys the independent and identically distributed normal distribution random variables; Y is the n×p response matrix; U is the n×l Y projection matrix; Q is the p×l orthogonal load matrix; matrix F is the error term, which obeys the independent and identically distributed normal distribution random variables.

[0068] Step 3.2: Perform data assimilation using the 4DVar cost function, taking into account the changes in observations over time scales. The assimilation formula is as follows:

[0069]

[0070] Among them, x k represents the model parameters, B is the model error, y is the number of observations, Q is the observation error, and H is the observation operator.

[0071] Step 3.3: Assign the data to the same spatial scale and apply it to the processed hyperspectral image data to identify whether the factor of cotton stress is pesticide damage or spider mite pests, and obtain the cotton pesticide damage distribution data and spider mite pest distribution data at the field scale.

[0072] In this embodiment, the method for constructing the hyperspectral recognition model for pesticide damage and spider mite pests includes:

[0073] 1) Obtain ground hyperspectral data;

[0074] 2) Use Savitzky-Golay to clean and denoise the data;

[0075] 3) Perform derivative and differential calculations;

[0076] The derivative is adopted and Calculation, Differentiation and ρ″(λ i )=[ρ′(λ i+1 )-ρ′(λ i-1 )] / 2(λ i+1 -λ i-1 ) to calculate and obtain the first-order derivative, second-order derivative, first-order differential, second-order differential, and fractional differential data of the spectral data

[0077] 4) Characteristic spectrum analysis: analyze the cotton canopy spectral data, derivative and differential data, screen sensitive spectrum segments, and extract characteristic spectrum segments;

[0078] 5) Based on the extracted characteristic spectra of cotton pesticide damage and spider mite pests, the partial least squares model was used to construct the recognition model.

[0079] In this embodiment, the ground object spectrometer is a SVC HR768 portable ground object spectrometer, with a spectrum range of 350nm-2500nm, 768 band channels, and a spectral resolution of 3.5nm@1000nm.

[0080] In this embodiment, the data collection requirements for the portable ground object spectrometer in step S22 to obtain cotton canopy spectral data from the cotton seedling stage to the flowering and boll stage are:

[0081] 1) The spectral data of each point of the sampling data shall be no less than 10 times;

[0082] 2) The ground visibility is not less than 10km; the amount of light cumulus clouds within a 90° solid angle around the sun should be less than 2%, with no cirrus clouds or cumulus congestus clouds; the wind force should be less than level 3;

[0083] 3) The portable ground object spectrometer uses a 25° field of view optical fiber probe to collect ground object spectral data.

[0084] A specific embodiment of the hyperspectral identification method for cotton pesticide damage and spider mite pests provided by the present invention:

[0085] The present invention is based on the territory of Shaya County, Aksu Prefecture, Xinjiang Uygur Autonomous Region. The area is located around the Tarim River. Cotton is the main crop grown locally, with a large scale and a simple planting structure. The seedling thrips and cotton aphids of cotton pests and diseases damage early. Continuous use of drugs in the seedling stage will accumulate and cause cotton to be seriously stressed by pesticides, while spider mite pests continue to occur during the same period. Data collection was carried out during the flowering and boll period of cotton from late June to late July. The cotton planting varieties were recommended varieties in Aksu Prefecture, and sowing began in mid-April. Spider mite stress cotton occurs naturally in the field, and pesticide stress cotton occurs due to the cumulative use of drugs by growers.

[0086] From late May to early August, in accordance with the "Technical Regulations for Integrated Prevention and Control of Major Cotton Diseases and Pests" (DB65T2271-2020), Beidou positioning equipment was used in the fields to accurately select high, medium and low 25m*25m sample plots, and collect cotton canopy spectral data in 1m*1m areas in the sample plots every 5 days, and record the growth of cotton, the number of diseases and pests, the degree of pesticide damage, etc.

[0087] Refer to the attached Figure 1 and attached Figure 2 , using a portable ground spectrometer to obtain cotton canopy spectral data from the seedling stage to the flowering and boll stage. The requirements for data collection are:

[0088] 1) The portable ground feature spectrometer uses the SVC HR768 model, with a spectrum range of 350nm-2500nm, 768 band channels, a spectral resolution of 3.5nm@1000nm, and a 25° field of view optical fiber probe to collect ground feature spectral data;

[0089] 2) The observation period is set at 10:30 to 17:30 local time to ensure sufficient solar altitude angle;

[0090] 3) Meteorological requirements are: ground visibility is not less than 10km; the amount of light cumulus clouds within a 90° solid angle around the sun should be less than 2%, with no cirrus clouds and cumulus congestus clouds, etc., and the wind force should be less than level 3;

[0091] 4) The spectral data of each point of the sampling data shall be no less than 10 times;

[0092] 5) According to the planting pattern of 12cm plant spacing, 66cm row spacing, and 12 cotton plants, the formula model is used to cover one row spacing. Among them, H is the height of the spectrometer probe from the canopy; L is the detection width of the ground object; θ is the field of view angle of the fiber optic probe. Calculate the height H of the spectrometer probe from the canopy, that is:

[0093] Assume that the row spacing is L = 66 cm and the field of view of the probe is 25°, so the height of the fiber optic probe from the cotton canopy is 150 cm;

[0094] Assume that the plant spacing is L = 12 cm and the field of view of the probe is 25°, so the height of the fiber optic probe from the cotton canopy is 28 cm;

[0095] The collected cotton canopy spectral data were preprocessed by using the Savitzky-Golay method, with the polynomial set to the fifth order, to perform denoising and smoothing on the vegetation spectral data to eliminate the influence of high-frequency noise.

[0096] The derivative and differential calculations are performed on the preprocessed data.

[0097] and Formula model for calculation;

[0098] Among them, R 1st (λ i ) is the spectral band λ i The first-order derivative value on R(λ i ) is the reflectivity value at band i, R(λ i+1 ) is the reflectivity value at the (i+1) band; R 2nd (λ i ) is the spectral band λ i The second-order derivative value on R(λ i ) is the reflectivity value at band i, R(λ i+1 ) is the reflectivity value at the (i+1) band;

[0099] Differentiation and ρ″(λ i )=[ρ′(λ i+1 )-ρ′(λ i-1 )] / 2(λ i+1 -λ i-1 ) formula model for calculation;

[0100] Among them, ρ′(λ i ) is the spectral band λ i The first-order differential value on ρ(λ i+1 ) is the reflectivity value at the (i+1) band; ρ(λ i-1 ) is the reflectivity value at the (i-1) band; ρ″(λ i ) is the spectral band λ i The second-order differential value on ρ′(λ i+1 ) is the first-order differential value at the (i+1) band; ρ′(λ i-1 ) is the first-order differential value at the (i-1) band; λ i is the wavelength value of band i; λ (i+1) is the wavelength value of the (i+1) band.

[0101] The first-order derivative, second-order derivative, first-order differential, second-order differential, and fractional differential (0.5 order, 1.5 order) data of the spectral data are obtained.

[0102] Calculate NDVI of canopy spectra (960，650) , using the formula model Calculate, where NDVI (960，650) is the normalized vegetation index; ρ 960 and ρ 650 is the reflectivity value of 960nm and 650nm band;

[0103] Red edge index R (550，760) Using formula model Calculate, where R (550，760) is the ratio vegetation index; ρ 760 and ρ 550 is the reflectivity value of 760nm and 550nm band;

[0104] GNDVI (760，550) The index adopts a formula model Calculate, where GNDVI (760，550) To improve the normalized difference vegetation index; ρ 760 and ρ 550 is the reflectivity value of 760nm and 550nm band;

[0105] The NNIR index uses a formula model Calculate, where NNIR is the vegetation index; ρ 760 , ρ 650 and ρ 550 It is the reflectivity value of 760nm, 650nm and 550nm bands;

[0106] The MCARI index uses the formula model MCARI = [(ρ 760 -ρ 650 )-0.2×(ρ 760 -ρ 550 )]×(ρ 760 -ρ 650 ) is calculated, where MCARI is the vegetation index; ρ 760 and ρ 650 It is the reflectivity value of 760nm and 650nm band;

[0107] The spectral indexes α and β of diseased plants were respectively based on the formula model Calculations and statistical analysis are performed to solve the optimal disease spectral parameters / index.

[0108] Using the red edge index R (550，760)Determine whether cotton has pests and diseases. If R (550，760) >0.2 can confirm that cotton is infected with spider mites; if R (550，760) When R is less than 0.13, it can be determined that cotton is initially infected with spider mites; (550，760) At the (0.13, 0.2) stage, a comprehensive judgment based on α and β is required: when α and β are greater than 6, the cotton plants are in a normal state; when α and β are less than 5.2, it indicates that the plants are in the process of transition from the early stage of infection to moderate disease.

[0109] Characteristic spectrum analysis: Analyze cotton canopy spectral data, derivative and differential data, and screen sensitive spectrum.

[0110] In the visible light 350nm-670nm spectrum, due to the absorption of chlorophyll in plant leaves, the canopy spectral reflectance of healthy cotton is low. As the degree of spider mite damage increases, the leaves of vegetation are damaged, and the reflectance is lower than that of healthy vegetation (such as the attached Figure 6 The reflectance of cotton plants under pesticide stress is higher than that of healthy plants, and as the degree of pesticide damage increases, a reflection peak appears in the red spectrum (670nm) (as shown in the attached figure). Figure 5 shown).

[0111] In the red edge 670nm-760nm spectrum, as the cotton pesticide damage and spider mite pests increase, the light moves toward the short-wave blue light, and a "blue shift" phenomenon occurs. The degree of "blue shift" of pesticide damage is greater than that of spider mite pests. The red edge index R (550,760) And NNIR index can be further used to distinguish diseases and spider mite pests.

[0112] In the near-infrared 760nm-1000nm spectrum, the reflectivity is higher than that of visible light and red edge spectrum due to the influence of plant cell tissue, water and other factors. With the occurrence of spider mite pests in plants, the plant tissue structure is damaged, the water content is reduced, and the reflectivity decreases as the severity of spider mite pests increases; while pesticide damage does not change the water content and physical structure, and the reflectivity is higher than that of normal cotton plants.

[0113] Therefore, in the 350nm-1000nm spectral range, characteristic spectral screening for cotton pesticide damage and spider mite pests can be carried out.

[0114] The partial least squares model was used to construct the recognition model based on the characteristic spectrum of cotton pesticide damage and spider mite pests.

[0115] Refer to the attached Figure 1 and attached Figure 3 A low-altitude UAV equipped with an imaging hyperspectral instrument was used to obtain hyperspectral images of the cotton canopy during the flowering and boll period. After data processing, the index calculation was completed.

[0116] As attached Figure 3As shown in the figure, after the vegetation index is calculated from the drone image, the vegetation area and non-vegetation area are first determined to filter out the vegetation area information;

[0117] In the vegetation area information (cotton and other vegetation), the cotton distribution area is identified according to the phenological calendar, image characteristics and vegetation index calculated from multiple images;

[0118] In the cotton distribution information, healthy cotton and unhealthy cotton are judged. If the cotton vegetation index and the spectral curve are consistent with the normal cotton state, it is normal cotton, otherwise it is unhealthy cotton (cotton with pesticide damage and pests and diseases).

[0119] In the unhealthy cotton information, using NDVI (960,650) , R (550,760) 、GNDVI (760,550) , NNIR, MCARI and α and β indices are used to comprehensively determine whether the cotton is damaged by pesticides and spider mites.

[0120] The UAV remote sensing platform of this embodiment includes: an eight-rotor UAV, a flight control system, a GaiaSky-mini2 airborne hyperspectral imager, a ground control system, a data processing system, and a micro portable computer HP Zbook 17G2. The eight-rotor UAV is designed and assembled by itself, with a single arm length of 65 cm, a fuselage weight of 8 kg, a load capacity of 12 kg, and a flight time of 30 minutes; the GaiaSky-mini2 airborne hyperspectral imager is a product of Sichuan Shuangli Hepu Company, which adopts a hovering scanning imaging method, with a spectral range of 400-1000nm, a spectral resolution of 3.5nm, a lateral field of view of 234m@500 meters, and a spatial resolution of 0.17m@500 meters.

[0121] The day of the UAV remote sensing operation was sunny and cloudless, with a wind speed of less than level 3, a speed of 4m / s, an altitude of 50m, a heading overlap of 75%, and a lateral overlap of 75%. After collecting data in accordance with regulations, the built-in Spec View software was used for data preprocessing.

[0122] The cotton canopy hyperspectral dataset of the experimental area was obtained. After processing with Spec View, Pix4D, ENVI, PIE and other software, a hyperspectral image with a ground resolution of 1 cm was obtained. After Savitzky-Golay smoothing and spectral differential processing, the standard preprocessed data was obtained and cropped using the ENVI ROI tool to obtain the hyperspectral image data of the target plot.

[0123] Refer to the attached Figure 1Based on the constructed hyperspectral recognition model of cotton pesticide damage and spider mite pests, data assimilation was performed by assimilation, the data were classified into the same spatial scale and applied to the processed hyperspectral image data to identify whether the factor of cotton stress is pesticide damage or spider mite pests, and the cotton pesticide damage distribution data and spider mite pest distribution data at the field scale were obtained.

[0124] After the spectral data and hyperspectral images were processed by the PCA (principal component analysis) method, the correlation between each factor was analyzed one by one, and the recognition model was constructed by fitting the partial least squares regression analysis model. The calculation formula is as follows:

[0125] X=TP T +E;

[0126] Y=UQ T +F;

[0127] Where X is the n×m prediction matrix; T is the n×l X projection matrix; P is the m×l orthogonal load matrix; Matrix E is the error term, which is an independent and identically distributed normal distribution random variable; Y is the n×p response matrix; U is the n×l Y projection matrix; Q is the p×l orthogonal load matrix; Matrix F is the error term, which is an independent and identically distributed normal distribution random variable

[0128] The 4DVar cost function is used for data assimilation, considering the changes in observations on the time scale, and using the formula model To assimilate,

[0129] Among them, x k represents the model parameters, B is the model error, y is the number of observations, Q is the observation error, and H is the observation operator.

[0130] The hyperspectral identification model of cotton pesticide damage and spider mite pests was applied to hyperspectral images to obtain the pesticide damage distribution data and spider mite pest distribution data at the field scale.

[0131] It will be apparent to those skilled in the art that the invention is not limited to the details of the exemplary embodiments described above and that the invention can be implemented in other specific forms without departing from the spirit or essential features of the invention. Therefore, the embodiments should be considered in all respects as exemplary and non-limiting, and the scope of the invention is defined by the appended claims rather than the foregoing description, and it is intended that all variations within the meaning and scope of the elements of the claims be included in the invention. Any reference numeral in a claim should not be considered as limiting the claim to which it relates.

[0132] As described above, the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features thereof may be replaced by equivalents. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A high-spectral identification method for cotton pesticide damage and spider mite pests, characterized in that: The steps include: Step 1: Obtain cotton canopy spectral data from the seedling stage to the flowering and boll stage through a ground object spectrometer and clean it, analyze the spectral characteristics of cotton pesticide damage and spider mite pests, and screen out cotton canopy spectral data of pesticide damage and spider mite pests; Step 2: Use a low-altitude UAV equipped with an imaging hyperspectral instrument to obtain hyperspectral images of the cotton canopy during the flowering and boll period. After data processing, complete the index calculation; Step 3: Construct a hyperspectral recognition model for pesticide damage and spider mite pests based on spectral features, apply the model to UAV hyperspectral images, perform feature determination, obtain the distribution map of pesticide damage and spider mite pests at the field scale, and output the final result; The method for constructing the hyperspectral recognition model for pesticide damage and spider mite pests in step 3 includes: 1) Obtain ground hyperspectral data; 2) Use Savitzky-Golay to clean and denoise the data; 3) Perform derivative and differential calculations, and the derivative is used; and Formula model for calculation; Among them, R 1st (λ i ) is the spectral band λ i The first-order derivative value on R(λ i ) is the reflectivity value at band i, R(λ i+1 ) is the reflectivity value at the (i+1) band; R 2nd (λ i ) is the spectral band λ i The second-order derivative value on R(λ i ) is the reflectivity value at band i, R(λ i+1 ) is the reflectivity value at the (i+1) band; Differentiation and ρ″(λ i )=[ρ′(λ i+1 )-ρ′(λ i-1 )] / 2(λ i+1 -λ i-1 ) formula model for calculation; Among them, ρ′(λ i ) is the spectral band λ i The first-order differential value on ρ(λ i+1 ) is the reflectivity value at the (i+1) band; ρ(λ i-1 ) is the reflectivity value at the (i-1) band; ρ″(λ i ) is the spectral band λ i The second-order differential value on ρ′(λ i+1 ) is the first-order differential value at the (i+1) band; ρ′(λ i-1 ) is the first-order differential value at the (i-1) band; λ i is the wavelength value of band i; (i+1) is the wavelength value of the (i+1) band; 4) Characteristic spectrum analysis: analyze the cotton canopy spectral data, derivative and differential data, screen sensitive spectrum segments, and extract characteristic spectrum segments; 5) Based on the extracted characteristic spectra of cotton pesticide damage and spider mite pests, the partial least squares model was used to construct the recognition model.

2. The method for high-spectrum identification of cotton pesticide damage and spider mite pests according to claim 1, characterized in that: In step 2, the UAV obtains the cotton canopy hyperspectral dataset of the test area, and after processing with Spec View, Pix4D, ENVI, and PIE software, a hyperspectral image with a ground resolution of 1 cm is obtained; after Savitzky-Golay smoothing and spectral differential processing are performed on the image, standard pre-processed data is obtained, and the ENVI ROI tool is used for cropping to obtain the hyperspectral image data of the target plot.

3. The method for high-spectrum identification of cotton pesticide damage and spider mite pests according to claim 1, characterized in that: The ground object spectrometer is a SVC HR768 portable ground object spectrometer with a spectrum range of 350nm-2500nm, 768 band channels, and a spectral resolution of 3.5nm@1000nm.

4. The method for high-spectrum identification of cotton pesticide damage and spider mite pests according to claim 1, characterized in that: The data collection requirements in step 1 are: 1) The spectral data of each point of the sampling data shall be no less than 10 times; 2) The ground visibility is not less than 10km; the amount of light cumulus clouds within a 90° solid angle around the sun should be less than 2%, with no cirrus clouds or cumulus congestus clouds; the wind force should be less than level 3; 3) The portable ground object spectrometer uses a 25° field of view optical fiber probe to collect ground object spectral data.

Images