Rice-stem borer pest stress identification method based on unmanned aerial vehicle hyperspectral image

Abstract

The invention relates to an unmanned aerial vehicle hyperspectral image-based rice-stem borer pest stress identification method, which comprises the following steps of: screening out wave bands sensitive to rice and pest rice by adopting an unmanned aerial vehicle hyperspectral image and a continuous projection algorithm, and constructing a rice index RI and a rice-stem borer pest index RBPI; thresholds of rice and non-rice are determined by analyzing a box plot of the RI index, thresholds of healthy rice and rice stressed by chilo suppressalis are determined by analyzing a box plot of the RBPI index, and an effective basis is provided for distinguishing sample types; according to the method, the decision tree algorithm is adopted to effectively identify the distribution of the rice stressed by the striped rice borer pests, the efficiency and accuracy of striped rice borer pest monitoring are improved, rapid monitoring of the striped rice borer pests is realized, an important scientific basis is provided for agricultural management and decision making, and green and sustainable development of agriculture is promoted.

Description

Technical Field

[0001] The invention belongs to the technical field of classification and extraction of remote sensing images, and in particular relates to a method for identifying rice stem borer pest stress based on unmanned aerial vehicle (UAV) hyperspectral images. Background Art

[0002] Rice is the foundation of my country's food security and occupies an extremely important position in my country's grain production. Its stable production and increased income are of great significance to ensuring national food security. However, rice is often threatened by the Chilo suppressalis pest. During the rice seedling stage, the larvae of the Chilo suppressalis invade the rice, causing the heart to wither; during the tillering stage, the larvae of the Chilo suppressalis continue to invade the rice stalks, causing the sheath to wither; from the jointing stage to the booting stage, the ears dry up and turn white, causing dead ears. Insect pests hinder the growth and development of rice, and in severe cases can lead to reduced yields or even total crop failure.

[0003] The traditional rice pest diagnosis method relies on producers and plant protection experts observing the changes in physiological and morphological indicators such as rice leaf color, leaf wilting, curling, and root health in the field to determine the severity of rice pest stress. This labor-intensive field survey method has problems such as low efficiency, high cost, and strong subjectivity.

[0004] Remote sensing technology can efficiently identify the spatial and temporal variation of crop pests. Currently, the remote sensing data commonly used for pest identification mainly include multispectral satellite data, hyperspectral satellite data (Landsat, Sentinel-2, Modis) and drone hyperspectral data. Multispectral and hyperspectral satellite data have low spatial resolution and long observation periods, and cannot capture small-scale pests in the field in a timely manner. Drones have high spatial resolution and flexible observation frequency, and can quickly cover large areas of farmland and accurately capture the occurrence of small-scale pests. In addition, drone hyperspectral images can provide rich spectral information, with each pixel having hundreds of bands of data. These data can be used to identify and distinguish different types of vegetation pests, which helps to detect pests earlier and provide a scientific basis for the implementation of prevention and control measures.

[0005] Obtaining the rice planting range is the basis for rice pest monitoring. Hyperspectral remote sensing often uses spectral indices to monitor rice planting areas. Common spectral indices include NDVI, EVI, DVI, etc. NDVI is mainly used to evaluate the density and greenness of vegetation, but it is easy to appear saturated in areas with high vegetation coverage (such as rice fields in the mature stage), and NDVI no longer increases significantly, making it difficult to effectively identify rice. EVI is insensitive to areas with mixed water bodies, but there are a lot of water bodies in the fields during the key phenological periods of rice growth (transplanting period, tillering period, and jointing period). In this case, EVI is easily affected by water bodies, resulting in inaccurate rice extraction results. DVI is highly sensitive to soil background and atmospheric conditions, but its application range is relatively limited. It is mainly used to distinguish between vegetation and non-vegetation, and it is difficult to further subdivide different vegetation. In addition, the above-mentioned vegetation index bands are wide and the spectra overlap, so that rice and other vegetation types have similar spectral reflectance characteristics in these bands, which makes it difficult to directly apply them to rice identification in drone hyperspectral images. Therefore, considering factors such as water mixing in rice fields, vegetation coverage, and similar spectral characteristics of different vegetation, it is necessary to develop a rice remote sensing index identification method for UAV hyperspectral images.

[0006] UAV hyperspectral remote sensing of rice pest monitoring also uses vegetation index methods, such as NDVI, GNDVI, SAVI, DVI, etc. They can all reflect the physiological state and health of vegetation and are more effective in identifying vegetation pests. NDVI is a commonly used index to measure vegetation greenness, which can reflect the uneven distribution of greenness on pest leaves. GNDVI is sensitive to the reduction of chlorophyll content and chlorophyll degradation. SAVI responds sensitively to the withering, curling or shedding of leaves of affected plants. DVI has a good monitoring capability for abnormalities such as yellowing, browning and spots on pest leaves.

[0007] However, the Chilo suppressalis mainly causes damage to the inside of rice stems, which is mostly manifested as dead core, dead sheath, and white panicle. The above vegetation index is difficult to effectively reflect the characteristics of rice stressed by Chilo suppressalis. Therefore, it is necessary to construct a spectral index of rice Chilo suppressalis pest stress to achieve effective monitoring of Chilo suppressalis pests. Summary of the invention

[0008] The purpose of the present invention is to overcome the deficiencies in the prior art and provide a method for identifying rice stem borer pest stress based on unmanned aerial vehicle hyperspectral images.

[0009] The rice stem borer pest stress identification method based on UAV hyperspectral imaging includes the following steps:

[0010] Step 1: Conduct a field survey of rice in the experimental area to obtain the land cover category and spatial location attributes; capture and obtain drone hyperspectral images;

[0011] Step 2: Determine the rice and non-rice sample points in the UAV hyperspectral image based on the field rice survey data, and perform spectral curve analysis on the obtained sample points;

[0012] Step 3: Using the continuous projection algorithm, the sensitive bands of rice and the sensitive bands of rice threatened by the rice stem borer are screened and identified from the data obtained from the field rice survey. The reflectance ratio of the sensitive bands is used to construct the rice identification index RI and the rice stem borer pest identification index RBPI, and thresholds are set for the two indexes respectively for judging the surface cover category.

[0013] Step 4: Use the decision tree algorithm to extract rice-growing areas from UAV hyperspectral images, and use the decision tree algorithm to identify rice areas threatened by the Chilo suppressalis within the rice-growing area; use the confusion matrix for accuracy evaluation.

[0014] Preferably, in step one, the categories of ground cover include rice and non-rice, and rice includes healthy rice and rice stressed by the striped stem borer; in step two, samples of non-rice areas and rice areas are distinguished according to the trend of reflectance changing with wavelength in the spectral curve, and samples of healthy rice and rice stressed by the striped stem borer are distinguished according to the reflectance value in the spectral curve.

[0015] As a preferred method, in step 3, a continuous projection algorithm is used to achieve dimensionality reduction of the entire band of 400 to 1000 nm. By cyclically calculating the projection of a wavelength on the remaining unselected wavelengths, the wavelength containing the least amount of redundant information is found, and the collinearity of the input data group is reduced by this wavelength, thereby screening out the two spectral wavelengths that are most sensitive to rice. The reflectances corresponding to the two spectral wavelengths from small to large are R S and R L , the calculation formula of rice identification index RI is:

[0016]

[0017] As a preferred method, in step 3, the thresholds T1 and R of the rice identification index RI are determined. L The threshold T2; when the sample RI>T1, and R L >T2, the surface cover category corresponding to the sample is rice; T1 is determined by calculating the pixel values ​​of the RI corresponding pixels of the rice and non-rice sample points in the UAV hyperspectral image, and making a box plot, and the threshold T1 is determined by the mean of the lower quartile of RI of the rice sample point and the upper quartile of RI of the non-rice sample point;

[0018] T2 is determined by calculating the R values ​​of rice and non-rice sample points in the UAV hyperspectral image. L , and make a box plot, by R of rice sample points LThe mean of the lower quartile and the upper quartile of the non-rice sample points determines the threshold T2 for R. L The mean of the upper quartile determines the threshold T2.

[0019] Preferably, in step three, the successive projections algorithm is used to reduce the dimension of the full wavelength band from 400 to 1000 nm. By circularly calculating the projection of one wavelength on the remaining unselected wavelengths, the wavelength with the least redundant information is found, and the collinearity of the input data set is reduced through this wavelength, so as to screen out the three spectral wavelengths that are most sensitive to rice stressed by Chilo suppressalis. The reflectivities corresponding to the three spectral wavelengths from small to large are R1, R2, and R3 respectively. The calculation formula for the rice Chilo suppressalis pest identification index RBPI is:

[0020]

[0021] Preferably, in step three, the threshold T3 for the rice Chilo suppressalis pest identification index RBPI is determined. If the RBPI of the sample with the surface cover category of rice is < T3, then the rice at this sample is healthy rice. The determination method of T3 is: calculate the pixel values of the RBPI corresponding pixels in the UAV hyperspectral image of the sample points of healthy rice and rice stressed by Chilo suppressalis, and the threshold T3 is obtained from the mean of the lower quartile of the RBPI of the healthy rice sample points and the upper quartile of the RBPI of the rice sample points stressed by Chilo suppressalis.

[0022] The beneficial effects of the present invention are:

[0023] 1) The present invention uses UAV hyperspectral images, and the successive projections algorithm is used to screen out the bands sensitive to rice and pest-infected rice, constructs the rice index RI and the rice Chilo suppressalis pest index RBPI. By analyzing the box plot of the RI index, the thresholds for rice and non-rice are determined, and by analyzing the box plot of the RBPI index, the thresholds for healthy rice and rice stressed by Chilo suppressalis are determined, thus providing an effective basis for the distinction of sample types.

[0024] 2) The present invention uses the decision tree algorithm to effectively identify the distribution of rice stressed by Chilo suppressalis pests, improves the efficiency and accuracy of Chilo suppressalis pest monitoring, realizes the rapid monitoring of rice Chilo suppressalis pests. This technology provides an important scientific basis for agricultural management and decision-making, and helps to promote the green and sustainable development of agriculture. Description of the Drawings

[0025] Figure 1 It is a sample point diagram of rice and non-rice, healthy rice and rice stressed by Chilo suppressalis;

[0026] Figure 2 It is a flow chart of the method for identifying rice stressed by Chilo suppressalis pests;

[0027] Figure 3 It is a schematic diagram of the ground object spectral curves of rice and non-rice;

[0028] Figure 4 This is a schematic diagram of the spectrum curve of the ground object of rice damaged by the Chilo suppressalis;

[0029] Figure 5 This is a schematic diagram of the screening of characteristic bands for rice;

[0030] Figure 6 This is a schematic diagram of the screening of characteristic bands for Chilo suppressalis;

[0031] Figure 7 The separation diagram of rice and non-rice by the index RI;

[0032] Figure 8 This is a diagram for separating rice and non-rice by reflectance at 770nm of the spectral band;

[0033] Fig. 9 This is a schematic diagram of rice identification results;

[0034] Fig.10 This is a diagram for separating healthy rice from rice infested with Chilo suppressalis using the RBPI index;

[0035] Fig.11 This is a schematic diagram of the identification results of rice stressed by Chilo suppressalis. DETAILED DESCRIPTION

[0036] The present invention is further described below in conjunction with embodiments. The description of the following embodiments is only used to help understand the present invention. It should be noted that for ordinary persons in the art, without departing from the principle of the present invention, the present invention can also be modified in some ways, and these improvements and modifications also fall within the scope of protection of the claims of the present invention.

[0037] Embodiment 1

[0038] As an embodiment, the rice stem borer pest stress recognition method based on UAV hyperspectral imaging is as follows: Figure 1 shown.

[0039] First, the rice stem borer pest stress index based on UAV hyperspectral images is provided, including:

[0040] Step 1: Conduct a field rice survey in the experimental area. Specific survey data include the categories and spatial location attributes of three types of surface cover: healthy rice, rice stressed by Chilo suppressalis, and non-rice.

[0041] Delineate the flight test area and obtain UAV hyperspectral images;

[0042] Preprocessing of UAV hyperspectral images includes radiation brightness calculation, reflectance correction, geometric correction, background value removal and mosaicking.

[0043] Step 2: Determine the sample points of rice and non-rice based on the field rice survey data, where rice includes healthy rice and rice stressed by the Chilo suppressalis; perform spectral curve analysis on the obtained sample points, and then obtain the spectral curve rules of rice and non-rice, healthy rice and rice affected by insect pests; specifically:

[0044] Rice and non-rice have their own unique spectral characteristics. The chlorophyll in rice leaves has a stronger absorption capacity for blue and red light, but less absorption for green light. Figure 3 As shown, the spectral curve of rice has low reflectivity at 400nm-510nm and 600nm-690nm, and a small reflection peak appears at 520nm-570nm. Due to the reflection of leaf cell structure, it fluctuates significantly at 700nm-781nm. The spectral reflectivity of non-rice areas changes relatively smoothly with wavelength, without obvious absorption or reflection peaks.

[0045] like Figure 4 As shown in the figure, in the spectral range of 400nm to 700nm, the reflectivity of healthy rice is lower than that of rice affected by insect pests. Specifically, the reflectivity of healthy rice plants is low in this visible light band because chlorophyll absorbs blue and red light strongly. The chlorophyll content of rice affected by insect pests decreases, resulting in a weakened light absorption capacity, so the reflectivity will be relatively high. In the near-infrared band of 730nm to 1000nm, the reflectivity of healthy rice is significantly higher than that of rice affected by insect pests. This is because healthy rice leaves have a good cell structure and a high water content, which can reflect near-infrared light more effectively. However, due to damaged leaves, disordered structure and reduced water content, the reflectivity of insect-infested rice in this band is low.

[0046] Step 3: Use the continuous projection algorithm to screen and identify the sensitive bands of rice, and use the sensitive bands to construct the rice identification index RI; draw a box plot to determine the RI threshold range.

[0047] The continuous projection algorithm was used to screen the sensitive bands of rice under the stress of Chilo suppressalis pests, and the rice Chilo suppressalis pest identification index (RBPI) was constructed; a box plot was drawn to determine the RBPI threshold range;

[0048] Step 4: Use the decision tree algorithm to extract rice-growing areas, and use the confusion matrix for accuracy evaluation; use the decision tree algorithm to identify rice areas under the threat of Chilo suppressalis pests within the rice-growing area, and use the confusion matrix for accuracy evaluation.

[0049] Embodiment 2

[0050] As another embodiment, this embodiment 2 proposes a more specific rice stem borer pest stress identification method based on drone hyperspectral images on the basis of embodiment 1:

[0051] According to the spectral differences between rice and non-rice, healthy rice and pest-infested rice obtained in step 2, the types of ground objects can be identified and distinguished. The spectral resolution of the drone hyperspectral image used reaches 18.5cm, which can capture these spectral differences in multiple bands. However, this high resolution brings the problem of band redundancy, increases the complexity of data processing, and leads to low computational efficiency. Therefore, it is necessary to reduce the dimensionality of the 272 bands of the drone hyperspectral image.

[0052] Specifically, in step three:

[0053] The successive projections algorithm (SPA) can achieve dimensionality reduction for the entire band of 400 to 1000 nm. The SPA algorithm calculates the projection of a wavelength on the remaining unselected wavelengths in a continuous cycle to find the wavelength with the least amount of redundant information, thereby reducing the collinearity of the input data set and screening out the spectral band that is most sensitive to rice. Figure 5 As shown, in the parameter setting, the number of characteristic wavelengths is set to 5, and 5 sensitive characteristic bands are screened out, and the wavelengths are 770nm, 510nm, 719nm, 541nm, and 784nm. According to the importance of the band characteristics, the first two most important bands are selected. In this embodiment, the wavelengths are 770nm and 510nm respectively.

[0054] Based on the sensitive bands and spectral characteristics of rice, an index RI for identifying rice is proposed to accurately identify the rice area. Affected by the chlorophyll, cell structure and water content of rice plants, rice has a higher reflectivity at 770nm and a relatively low reflectivity at 510nm. In contrast, non-rice has a lower reflectivity at 770nm and a higher reflectivity at 510nm. Therefore, by calculating the reflectivity ratio of 770nm to 510nm, the difference between rice and non-rice can be highlighted. In addition, in order to better remove the influence of weeds on rice extraction, the reflectivity of the 770nm band is added to further distinguish between rice and non-rice. Figure 8 As shown, R 770 The addition of expands the difference between rice and non-rice, which can better realize the identification and monitoring of rice. The calculation formula of index RI is:

[0055]

[0056] Among them, R 770 is the reflectance value of the spectral band at 770nm, R 510 is the reflectance value of the spectral band at 510nm. When the sample's RI>T1, and R 770 >T2, the corresponding land cover category of the sample is rice, and T1 is The threshold value of T2 is R 770 The threshold value.

[0057] The threshold T1 is determined by calculating the number of rice and non-rice sample points in the UAV hyperspectral image. The pixel values ​​of the corresponding pixels are plotted and a box plot is made. The threshold T1 is determined by the mean of the lower quartile of the rice sample points and the upper quartile of the non-rice sample points.

[0058] The threshold T2 is determined by calculating the reflectance of rice and non-rice sample points in the 770nm band and making a box plot. The threshold T2 is determined by the mean of the lower quartile of rice sample points and the upper quartile of non-rice sample points.

[0059] Based on the same method, SPA feature selection was performed on the spectral data of rice sensitive to Chilo suppressalis stress, such as Figure 6 As shown in the figure, the number of characteristic wavelengths is set to 5, and 5 sensitive characteristic bands are screened out, with wavelengths of 781nm, 724nm, 510nm, 464nm, and 688nm. According to the importance of band characteristics, the top 3 most important bands are selected, with wavelengths of 781nm, 724nm, and 510nm.

[0060] Based on the spectral bands that rice pests are sensitive to, a rice stem borer pest index (RBPI) is proposed. Fig.10 As shown in the figure, the RBPI of healthy rice is higher than that of insect-infested rice, which can effectively distinguish healthy rice from rice stressed by the Chilo suppressalis. The calculation formula of the index RBPI is:

[0061]

[0062] Among them, R 781 , R 724 , R 510 It is the reflectance value of the spectral band at 781nm, 724nm, and 510nm. When the RBPI of the sample with the surface cover category of rice is greater than T3, the rice in the sample is under the stress of Chilo suppressalis, and T3 is the threshold of RBPI.

[0063] Draw a box plot to determine the RBPI threshold range. Calculate the pixel values ​​of the RBPI corresponding pixels in the UAV hyperspectral image for healthy and striped stem borer type sample points, and make a box plot to determine the threshold T3. The threshold T3 is determined by the mean of the lower quartile of the healthy sample points and the upper quartile of the striped stem borer sample points.

[0064] It should be noted that the parts in this embodiment that are the same or similar to those in the first embodiment can be referenced to each other and will not be described in detail in this application.

[0065] Embodiment 3

[0066] As another embodiment, this embodiment 3 proposes, based on the first and second embodiments, that the rice stem borer pest stress identification method based on the UAV hyperspectral image uses a rice stem borer pest extraction device, including:

[0067] The acquisition module is used to acquire the UAV hyperspectral image data and pre-process the UAV hyperspectral data;

[0068] The screening module is used to screen the characteristic sensitive bands from 272 hyperspectral bands;

[0069] Building modules for constructing the rice index RI and rice stem borer pest index RBPI based on UAV hyperspectral technology;

[0070] Index calculation module, used to calculate the corresponding index RI and RBPI image of drone hyperspectral image;

[0071] The extraction module is used to determine the threshold range according to the RI result of the sample to extract the rice range, and then determine the threshold range according to the RBPI result to extract the rice stem borer range.

[0072] A computer storage medium is also provided, in which a computer program is stored; when the computer program is run on a computer, the computer executes any rice stem borer pest identification method based on unmanned aerial vehicle hyperspectral imagery as described in the first aspect.

[0073] The rice planting area was extracted by the decision tree algorithm, and the confusion matrix was used for accuracy evaluation. The quantitative evaluation indicators include producer accuracy (PA), user accuracy (UA), overall classification accuracy (OA) and Kappa. The calculation formulas of the evaluation indicators are:

[0074]

[0075] Where TP is the true positive of a certain class and FN is the false negative.

[0076]

[0077] Where TP is a true positive and FP is a false positive.

[0078]

[0079] In the formula, ∑TP represents the sum of true positive samples of all categories, and ∑N is the total number of samples.

[0080]

[0081] Where OA is the overall accuracy, P eis the accuracy obtained by random classification, N represents the total number of validation samples, n represents the total number of columns in the confusion matrix, xii represents the number of samples in the ith row and ith column of the confusion matrix, and X i+ and X +i Respectively represent the sum of the elements in each row and column of the confusion matrix.

[0082] In this embodiment, a decision tree algorithm is used to divide the rice planting range using the RI index threshold. Among them, the producer accuracy (PA) of rice is 0.96 and the user accuracy (UA) is 0.97; the PA and UA of non-rice are both 0.98, the overall classification accuracy (OA) is 0.97, and the Kappa is 0.92. After evaluation, the recognition results show high accuracy and good performance in the recognition of rice and non-rice.

[0083] Using the same index, in the rice-growing area, the range of rice stressed by Chilo suppressalis was extracted through RBPI threshold division, and the confusion matrix was used for accuracy evaluation. Among them, the PA of rice stressed by Chilo suppressalis was 0.96, and the UA was 0.95; the PA of healthy rice was 0.94, and the UA was 0.95. The overall classification accuracy (OA) was 0.95, and Kappa was 0.9. The overall accuracy of the recognition was high, showing that the index method used can effectively distinguish between rice stressed by Chilo suppressalis and healthy rice, with high accuracy.

[0084] It should be noted that the parts in this embodiment that are the same or similar to those in Embodiments 1 to 2 can be referenced to each other and will not be described in detail in this application.

[0085] Embodiment 4

[0086] As another embodiment, this embodiment 4 proposes a more specific rice stem borer pest stress identification method based on UAV hyperspectral images on the basis of embodiments 1 to 3:

[0087] In step 1, use the polygon tool to delineate the collection area and flight area.

[0088] Set the parameters of the drone collection area, including the color of the reflectivity calibration cloth, flight altitude and speed. Use white cloth as the reflectivity calibration cloth to ensure that the area of ​​the calibration cloth is not overexposed, that is, the peak value of the pixel brightness value (DN value) curve is less than 4000, and at the same time ensure that the gaze time per frame is greater than the exposure time. Set the flight altitude to 200m and calculate the flight speed using the software's built-in calculator.

[0089] The flight area plans the route, the flight altitude matches the altitude setting of the collection area, which is 200m, the lateral overlap rate is 30%, the heading overlap rate is 10%, the margin is 0, and the main route angle is adjusted to be parallel to the long side of the collection area.

[0090] In step 1, the process of preprocessing the UAV hyperspectral image is as follows:

[0091] The radiance calculation is to convert the DN data collected by the hyperspectral imager into radiance data through a standard radiance calibration configuration file. The specific conversion formula is:

[0092]

[0093] Where DN is the original light intensity value recorded by the hyperspectral imager, L1 is the radiation brightness value, G is the camera gain, texp is the integration time of the photosensitive element, and DF is the dark current noise.

[0094] The reflectivity correction first generates a white reference file based on the pixel reflectivity (0.56) of the reflectivity white cloth, and then calculates the reflectivity of all strip data in the acquisition area.

[0095] Geometric correction uses a geometric correction algorithm to automatically match each frame of data from the hyperspectral imager with the corresponding attitude information, and completes the geometric correction after a successful match.

[0096] Background value removal and mosaicking. Add the value of ignoring zero radiation brightness to each geometrically corrected txt file to complete the background value removal, and then use the batch fast mosaicking tool to complete the image mosaicking. After preprocessing, the drone hyperspectral data is output with 272 bands and a spatial resolution of 18.5cm.

[0097] It should be noted that the parts in this embodiment that are the same or similar to those in Embodiments 1 to 3 can be referenced to each other and will not be described in detail in this application.

[0098] The various embodiments in this specification are described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the various embodiments can be referenced to each other.

Claims

1. A method for identifying rice stem borer pest stress based on unmanned aerial vehicle hyperspectral imaging, characterized in that: It includes the following steps: Step 1: Conduct on-site rice surveys in the experimental area to obtain surface cover categories and spatial location attributes; capture and obtain UAV hyperspectral images; Step 2: Determine the sample points of rice and non-rice in the UAV hyperspectral images based on the on-site rice survey data, and conduct spectral curve analysis on the obtained sample points; Step 3: Use the successive projections algorithm to screen and identify the sensitive bands of rice and the sensitive bands of rice under the stress of Chilo suppressalis from the data obtained from the on-site rice survey. Construct the rice recognition index R and the rice Chilo suppressalis pest recognition index RBPI using the reflectance ratios of the sensitive bands, and set thresholds for the two indices respectively for the judgment of surface cover categories; Step 4: Use the decision tree algorithm to extract the rice planting area from the UAV hyperspectral images, and use the decision tree algorithm to identify the rice areas under the stress of Chilo suppressalis within the rice planting range; use the confusion matrix for accuracy evaluation.

2. The rice stem borer pest stress identification method based on unmanned aerial vehicle hyperspectral imaging according to claim 1 is characterized in that: In Step 1, the categories of surface cover include rice and non-rice, and rice includes healthy rice and rice under the stress of Chilo suppressalis; In Step 2, distinguish the samples of non-rice areas and rice areas according to the trend of reflectance changing with wavelength in the spectral curve, and distinguish the samples of healthy rice and rice under the stress of Chilo suppressalis according to the reflectance values in the spectral curve.

3. The rice stem borer pest stress identification method based on unmanned aerial vehicle hyperspectral imaging according to claim 1 is characterized in that: In step 3, the continuous projection algorithm is used to achieve dimensionality reduction of the entire band of 400 to 1000 nm. By cyclically calculating the projection of a wavelength on the remaining unselected wavelengths, the wavelength containing the least redundant information is found. This wavelength is used to reduce the collinearity of the input data set, thereby screening out the two spectral wavelengths that are most sensitive to rice. The reflectances corresponding to the two spectral wavelengths from small to large are R S and R L , the calculation formula of rice identification index RI is:

4. The rice stem borer pest stress identification method based on unmanned aerial vehicle hyperspectral imaging according to claim 3 is characterized in that: In step 3, the thresholds T1 and R of the rice identification index R are determined. L The threshold T2; when the sample RI>T1, and R L >T2, the surface cover category corresponding to the sample is rice; T1 is determined by calculating the pixel values ​​of the RI corresponding pixels of the rice and non-rice sample points in the UAV hyperspectral image, and making a box plot, and the threshold T1 is determined by the mean of the lower quartile of RI of the rice sample point and the upper quartile of RI of the non-rice sample point; T2 is determined by calculating the R values ​​of rice and non-rice sample points in the UAV hyperspectral image. L , and make a box plot, by R of rice sample points L R of the lower quartile and non-rice sample points L The mean of the upper quartile determines the threshold T2.

5. The rice stem borer pest stress identification method based on unmanned aerial vehicle hyperspectral imaging according to claim 1, characterized in that: In Step 3, use the successive projections algorithm to achieve dimensionality reduction of the full band from 400 to 1000 nm. By circularly calculating the projection of one wavelength on the remaining unselected wavelengths, find the wavelength with the least redundant information content, and reduce the collinearity of the input data group through this wavelength, so as to screen out the three spectral wavelengths most sensitive to rice under the stress of Chilo suppressalis. The reflectances corresponding to the three spectral wavelengths from small to large are R1, R2, and R3 respectively. The calculation formula for the rice Chilo suppressalis pest recognition index RBPI is:

6. The rice stem borer pest stress identification method based on unmanned aerial vehicle hyperspectral imaging according to claim 5 is characterized in that: In Step 3, determine the threshold T3 of the rice Chilo suppressalis pest recognition index RBPI. When RBPI of the sample with the surface cover category of rice < T3, the rice at this sample is healthy rice; the determination method of T3 is: calculate the pixel values of the RBPI corresponding pixels of the sample points of healthy rice and rice under the stress of Chilo suppressalis in the UAV hyperspectral images, and obtain the threshold T3 from the mean value of the lower quartile of RBPI of the healthy rice sample points and the upper quartile of RBPI of the rice sample points under the stress of Chilo suppressalis.

Images