A crop disease and pest monitoring method based on unmanned aerial vehicle remote sensing technology

Abstract

The application discloses a kind of construction methods of crop disease and insect pest monitoring model, comprising the following steps:1) unmanned aerial vehicle multi-view multispectral image data acquisition is carried out to crop canopy;2) the unmanned aerial vehicle multi-view multispectral image data collected is respectively preprocessed, including image stitching, geometric correction, radiation correction and canopy extraction;3) random forest algorithm is used to extract crop area;4) the crop area extracted in step 3) is based on the band sensitive to crop disease and insect pest, and the different generation combination is used to calculate the spectral index sensitive to disease and insect pest;5) random forest algorithm is used, and the importance of each spectral index of different observation angles in step 4) in disease and insect pest monitoring is evaluated using average drop gini coefficient;6) forward selection algorithm is used to determine the optimal spectral index combination of different observation angles;7) random forest algorithm is used to construct crop disease and insect pest monitoring model.

Description

TECHNICAL FIELD

[0001] The present application relates to a crop disease and pest monitoring method, in particular to a crop disease and pest monitoring method based on unmanned aerial vehicle remote sensing technology. BACKGROUND

[0002] In the process of agricultural production, the occurrence of crop diseases and pests often causes huge yield and economic losses, and seriously threatens food security. The traditional investigation method is manual inspection, which is time-consuming and labor-intensive. The emergence and rapid development of unmanned aerial vehicles can obtain high spatial resolution multispectral images in near real time, providing an unprecedented opportunity for crop disease and pest monitoring in the field. Compared with traditional methods, unmanned aerial vehicle remote sensing survey has many advantages. Unmanned aerial vehicle remote sensing survey can quickly and efficiently cover a large area and provide a comprehensive field of view.

[0003] Unmanned aerial vehicle remote sensing survey can also detect signs of diseases and pests at an early stage to allow timely intervention. In addition, unmanned aerial vehicles equipped with multispectral sensors can detect changes in plant health that are invisible to the naked eye, allowing more accurate assessment of the impact of plant diseases and pests.

[0004] Patent application CN202010392669.5 discloses a grid-based rice sheath blight habitat evaluation method based on comprehensive multi-source remote sensing information.

[0005] However, the shielding effect of the plant canopy in the field hinders the use of unmanned aerial vehicle observation technology for field disease and pest monitoring, posing a serious challenge to unmanned aerial vehicle remote sensing survey. SUMMARY

[0006] In order to overcome the shortcomings of the above-mentioned prior art, the purpose of the present application is to provide a crop disease and pest field monitoring method using plant canopy multi-view multispectral imaging technology.

[0007] In order to achieve the above-mentioned purpose, the present application provides a crop disease and pest monitoring method based on unmanned aerial vehicle multi-view remote sensing technology, comprising the following steps:

[0008] 1) unmanned aerial vehicle multi-view multispectral image data acquisition of the crop canopy in the field;

[0009] 2) image stitching, geometric correction, radiation correction and canopy extraction and other preprocessing of the collected unmanned aerial vehicle multi-view multispectral image data;

[0010] 3) using a random forest algorithm to build a model to extract crop areas from unmanned aerial vehicle multispectral images;

[0011] 4) based on the wavebands sensitive to crop diseases and pests, using different combinations of different generations to calculate the spectral indices sensitive to diseases and pests;

[0012] 5) Using random forest algorithm, the importance of different observation angles of spectral indices in disease and pest monitoring is evaluated by average descending Gini coefficient.

[0013] 6) The optimal combination of spectral indices of different observation angles is determined by forward selection method.

[0014] 7) The crop disease and pest monitoring model is constructed by random algorithm.

[0015] Further, in the above step 1), the unmanned aerial vehicle multi-view multispectral image data can be the reflectance spectral imaging data of the crop canopy within the observation angle range of 0-90°.

[0016] Further, in the above step 1), the unmanned aerial vehicle multi-view multispectral image data can be the reflectance spectral imaging data of the crop canopy within the observation angle range of 0-30°.

[0017] Further, in the above step 1), the unmanned aerial vehicle multi-view multispectral image data can be the reflectance spectral imaging data of the crop canopy within the wavelength range of 350-900 nm.

[0018] Further, in the above step 1), the unmanned aerial vehicle multi-view multispectral image data can be the reflectance spectral imaging data of the crop canopy within the wavelength range of 400-870 nm.

[0019] Further, in the above step 1), the plant sample of the field crop can be a healthy field crop plant sample and a field crop plant sample at different disease stages.

[0020] Further, in the above step 2), image stitching is to align and stitch the unmanned aerial vehicle multispectral images collected at different observation angles.

[0021] Further, in the above step 2), the radiation correction is to construct an empirical relationship between the known band reflectance provided by the standard diffuse reflectance plate and its pixel value in the original image, and then perform radiation correction to obtain the reflectance data of the multispectral image.

[0022] Further, in the above step 3), the random forest algorithm is used to construct the model, extract the crop area from the unmanned aerial vehicle multispectral image, classify the bare soil and weeds in the image as non-crop area, and obtain the binary classification mask map of crop / non-crop, which is used for extraction and calculation of disease and pest sensitive spectral indices.

[0023] Further, in the above step 4), the spectral index can be a spectral index closely related to various physiological and biochemical characteristics of the leaf.

[0024] Further, in the above step 5), the average descending Gini coefficient calculated by the random forest algorithm is used to evaluate the importance of the spectral index of different observation angles in crop disease and pest monitoring.

[0025] Further, in the above step 6), based on the calculated spectral index importance, a forward selection method is used to determine the optimal spectral index combination of different observation angles, and the greater the average descending Gini coefficient value of the input features, the more important the input features.

[0026] Further, in the above step 6), the spectral index observation data of multiple angles can be fused, and the random forest algorithm can be used to effectively avoid the influence of multicollinearity by using its sample randomness and variable randomness, so as to construct a crop disease and pest monitoring model based on multi-view multi-spectral observation data fusion.

[0027] Further, the multi-angle unmanned aerial spectral index of the crop to be tested is input into the detection model constructed in the above step 7), so that rapid and non-destructive detection of crop diseases and pests can be realized.

[0028] Further, the model constructed in step 7) can complete the detection calculation of crop diseases and pests at the field scale within a few seconds.

[0029] Specifically, the present application is as follows:

[0030] 1. A method for constructing a crop disease and pest monitoring model, comprising the following steps:

[0031] 1) collecting unmanned aerial multi-view multi-spectral image data of crop canopies;

[0032] 2) respectively pre-processing the collected unmanned aerial multi-view multi-spectral image data, including image stitching, geometric correction, radiation correction and canopy extraction;

[0033] 3) extracting crop regions from the pre-processed unmanned aerial multi-view multi-spectral image data in step 2) using a random forest algorithm;

[0034] 4) based on the wavebands sensitive to crop diseases and pests, calculating spectral indices sensitive to diseases and pests by using different combination of generations for the crop regions extracted in step 3);

[0035] 5) using a random forest algorithm to evaluate the importance of each spectral index of different observation angles in step 4) in disease and pest monitoring by using the average descending Gini coefficient;

[0036] 6) determining the optimal spectral index combination of different observation angles using a forward selection algorithm;

[0037] 7) Constructing a crop disease and pest monitoring model using a random forest algorithm.

[0038] 2. The method of claim 1, wherein the multi-angle image data of the unmanned aerial vehicle in step 1) is multi-angle multi-spectral image data of the crop canopy collected at an observation angle of 0-90°.

[0039] 3. The method of claim 2, wherein the multi-angle image data of the unmanned aerial vehicle in step 1) is multi-angle multi-spectral image data of the crop canopy collected at an observation angle of 0-30°.

[0040] 4. The method of any one of claims 1-3, wherein the multi-angle image data of the unmanned aerial vehicle in step 1) is multi-angle multi-spectral image data of the crop canopy in the wavelength range of 350-900 nm.

[0041] 5. The method of claim 4, wherein the multi-angle image data of the unmanned aerial vehicle in step 1) is multi-angle multi-spectral image data of the crop canopy in the wavelength range of 400-870 nm.

[0042] 6. The method of any one of claims 1-5, wherein in step 3), bare soil and weeds in the image are classified as non-crop areas to obtain a binary classification mask of crop / non-crop, which is used for extraction and calculation of disease and pest sensitive spectral indices.

[0043] 7. The method of any one of claims 1-6, wherein the spectral indices sensitive to disease and pests in step 4) are spectral indices closely related to various physiological and biochemical characteristics of the crop.

[0044] 8. The method of any one of claims 1-7, wherein in step 5), the average decline Gini coefficient is used to quantify the influence of heterogeneity information observed by the input feature decision tree node, so as to reflect the importance of the input feature.

[0045] 9. The method of claim 1 to 8, wherein in step 6), the inputted spectral indices for determining the optimal combination of spectral indices at different observation angles include spectral indices at different observation angles and spectral indices fused from multiple observation angles.

[0046] 10. The method of claim 1 to 9, wherein in step 6), the forward selection algorithm starts from an empty model and iteratively adds spectral indices to the model in order of importance to maximize the model performance over the previous step.

[0047] 11. The method of claim 10, wherein in step 6), the performance of each step of model construction is evaluated based on the accuracy of the random forest model, and the forward selection process continues to add the most important spectral indices one by one until all indices are considered.

[0048] 12. The method of claim 1 to 11, wherein the crop disease and pest is corn southern rust.

[0049] 13. Use of the method of claim 1 to 10 in crop disease and pest monitoring or control.

[0050] The present application provides an efficient, non-destructive, repeatable and high-precision detection method, which provides valuable information for detecting and timely preventing crop diseases and pests. Compared with the prior art, the present application uses five spectral channels captured by the unmanned aerial vehicle in the crop field: blue light, green light, red light, red edge and near infrared, and calculates various spectral enhancement specific features based thereon, fuses multi-view data, and establishes a crop disease and pest monitoring model at the field scale. Using the crop disease and pest monitoring model of the present application, the crop disease and pest can be quickly and non-destructively identified in the field.

[0051] Unmanned aerial vehicle multi-view multi-spectral observation can effectively overcome the shielding of crop canopy, collect spectral information at different layers of the canopy, and provide a feasible path for improving crop disease and pest monitoring at the field scale. Multi-view multi-spectral observation of the canopy based on the unmanned aerial vehicle can collect complementary spectral information at different layers of the crop, improve the utilization efficiency of data, and improve the monitoring accuracy of diseases and pests, providing a reference for crop disease and pest monitoring and control, and having great application potential in crop disease and pest monitoring in the field. BRIEF DESCRIPTION OF DRAWINGS

[0052] Figure 1 The technical flowchart of an embodiment of the present application.

[0053] Figure 2 Importance ranking of spectral indices for different observation angles (0°, 15°, 30°) of one embodiment of the present application in corn southern rust monitoring.

[0054] Figure 3 Stepwise results of forward selection algorithm based on random forest scores (OA) for one embodiment of the present application.

[0055] Figure 4 Overall accuracy and F1 score of different corn southern rust monitoring models for one embodiment of the present application.

[0056] Figure 5 Field corn southern rust monitoring mapping for one embodiment of the present application, where the left image is the field high-resolution image, and the right image is the corn southern rust intensity based on the multi-view optimal model. DETAILED DESCRIPTION

[0057] In order to better understand the present application, the following examples are given in conjunction with the accompanying drawings. It should be understood that the examples of the present application are only used to explain the present application and are not limited to the present application. The scope of protection of the present application is only limited by the claims of the present application. The examples provided by the present application are only some preferred examples, and do not limit the present application in any form. Those skilled in the art can make changes, equivalent substitutions or modifications according to the content of the present application to form different embodiments. However, any changes and modifications to the method of the present application, any equivalent substitutions, are within the scope of protection of the present application, without departing from the concept of the present application.

[0058] The present application discloses a crop disease and pest monitoring method based on unmanned aerial vehicle multi-view remote sensing technology. The method of the present application is suitable for the fields of crop health diagnosis, crop disease and pest control, and pesticide quantitative spraying technology. The present application can establish a crop disease and pest monitoring model according to multi-view remote sensing data of crops combined with a variety of vegetation indices, and rapidly and non-destructively identify crop diseases and pests in the field.

[0059] Figure 1A technical flowchart of one embodiment of the present application. First, the unmanned aerial vehicle multi-view multi-spectral image data of the crop canopy is collected. Next, the collected image data is spliced, geometrically corrected, radiometrically corrected and canopy extracted. The disease condition is investigated in the experimental plot, and the RTK (Real-time kinematic) positioning technology of sub-meter level is used to mark and locate the investigation sample points. Through the comprehensive calculation of the disease degree, the intensity of the disease and insect pests is determined. Next, the random forest algorithm is used to extract the crop area, and based on the sensitive band of the crop disease and insect pests, the spectral index sensitive to the disease and insect pests is calculated. Next, the average descending Gini coefficient is used to evaluate the importance of each spectral index in the disease and insect pest monitoring, and the forward selection algorithm is used to determine the optimal spectral index combination of different observation angles. Next, the random forest algorithm is used to construct the disease and insect pest monitoring model of the crop single angle and multi-angle fusion, the model precision evaluation comparison is carried out, and finally the field disease is obtained.

[0060] Next, taking corn southern rust as an example, the corn southern rust monitoring model is constructed by using unmanned aerial vehicle multi-spectral image. On the basis of sensitive spectral index combination optimization, the random forest algorithm is used to screen the improvement index of corn southern rust monitoring by using multi-view multi-spectral observation data fusion mechanism, and the corn southern rust monitoring model is constructed, and then the observation data is input into the optimal model to realize the mapping of field disease.

[0061] Example 1, unmanned aerial vehicle multi-angle multi-spectral image data collection

[0062] The DJI Phantom 4 Multispectral (P4M) unmanned aerial vehicle produced by Shenzhen DJI Innovation Technology Co., Ltd. was used to obtain multi-angle spectral data of the corn canopy. In order to maintain consistency with the ground data, the unmanned aerial vehicle data was collected between 11:00 and 14:00 on March 12-13, 2021, when the weather conditions were good and sunny with few clouds. After considering the accuracy of the image and the efficiency of image acquisition, the flight parameters were set as follows: height 30 m, lateral overlap rate 80%, forward overlap rate 75%, and flight speed 2.3 m / s. Before executing the flight task, several ground control points were laid out in the field, and the coordinate information of the ground control points was recorded using RTK positioning technology to establish the geographic reference of the unmanned aerial vehicle image. At the same time, three standard diffuse reflectance plates were placed in the field for radiation correction of the obtained unmanned aerial vehicle images. By adjusting the gimbal angle, multi-spectral images of the unmanned aerial vehicle were obtained at 0° (horizontal view), 15°, and 30° view angles. The spatial resolution of the obtained images was 1.5 cm, the spectral range was 400-870 nm, and it included 5 channels, namely blue light (450±16 nm), green light (560±16 nm), red light (650±16 nm), red edge (730±16 nm), and near-infrared (840±26 nm) bands.

[0063] Example 2, Field Data Collection of Pests and Diseases

[0064] The severity of corn southern rust was investigated in the experimental field plot. A 1 m x 1 m plot was set up at each survey point, and the severity of corn southern rust in the plot was determined by a comprehensive calculation of the disease severity of the corn plants in the plot. To ensure that the survey points correspond accurately to the images collected by the unmanned aerial vehicle in terms of spatial location, the survey points were marked and located using sub-meter RTK positioning technology, which had a horizontal error of 0.01 m and a vertical error of 0.02 m in the field environment.

[0065] Example 3, Preprocessing of Unmanned Aerial Vehicle Multi-View Multi-Spectral Image Data

[0066] DJI Smart Editor was used to align and stitch the collected unmanned aerial vehicle multi-spectral images at different observation angles. During the two-dimensional reconstruction process, the ground control points were imported for geometric correction to further improve the horizontal and vertical accuracy of the image reconstruction. In addition, the known band reflectance provided by the standard diffuse reflectance plate was used to construct an empirical relationship with its DN (Digital Number) value in the original image to perform radiation correction and obtain the reflectance data of the multi-spectral image.

[0067] Example 4, Extraction of Crop Canopy Region

[0068] A stratified strategy was adopted to distinguish between corn and non-corn areas. First, high-resolution UAV imagery was used to differentiate between crops and non-crops. Then, pest and disease monitoring was conducted on the crops, which helps improve monitoring accuracy. A random forest algorithm was used to construct a model to extract corn areas from the UAV multispectral imagery. Bare soil and weeds in the imagery were classified as non-corn areas, resulting in a binary classification mask map for corn / non-corn.

[0069] Example 5: Calculation of the spectral sensitivity index for pests and diseases

[0070] The most sensitive bands for maize southern rust, including red, green, blue, red-edge (RE), and near-infrared (NIR), were selected, and spectral indices for sensitivity to maize southern rust were calculated using different algebraic combinations. The 15 selected spectral indices for sensitivity to the disease are shown in Table 1.

[0071] Table 1: 15 Pest-Specific Spectral Indices Selected for Southern Rust Detection of Maize in this Invention

[0072]

[0073] Note: Red, Green, Blue, RE, and NIR refer to the reflectivity of the red band, green band, blue band, red edge band, and near-infrared band, respectively.

[0074] Example 6: Screening the spectral indices most suitable for constructing pest and disease detection models

[0075] First, the average decreasing Gini coefficient calculated using the random forest algorithm is used to assess the importance of spectral indices from different observation angles in the monitoring of southern rust in maize. The average decreasing Gini coefficient quantifies the impact of input features on the heterogeneous information observed at decision tree nodes, thus reflecting the importance of the input features; a larger average decreasing Gini coefficient value indicates a more important input feature.

[0076] See results Figure 2 , Figure 2 This paper ranks the importance of spectral indices at different observation angles (0°, 15°, 30°) in monitoring southern rust of maize according to an embodiment of the present invention. The results show that the GPRI index, calculated based on the red-green band ratio, is the most important feature for monitoring southern rust of maize at all observation angles, followed by the NPCI index, calculated based on red-green band normalization. Both red and green bands are closely related to the pigment content and photosynthesis of plants.

[0077] Example 7: Determining the optimal combination of spectral features for different observation angles

[0078] Based on the calculated spectral index importance, a forward selection method was used to determine the optimal spectral feature combination at different observation angles. The forward selection algorithm starts from an empty model, iteratively adds spectral indices to the model according to the importance of the model performance, and maximizes the improvement of the previous step. The performance of each step is evaluated according to the accuracy of the model (i.e. random forest), and the forward selection process continues to add the "best" spectral features (i.e. the most important spectral features) one by one until all indices are considered. The optimal spectral feature combination at different observation angles determined by the forward selection algorithm is used as the input of the model.

[0079] The optimal spectral feature combination selected by the forward selection algorithm is shown in FIG. 4. Figure 3 Figure 3 The step-by-step results of the forward selection algorithm based on the random forest score (OA). The results show that after gradually adding spectral features from an empty model, the accuracy of the model shows a significant increase, and after a certain step, the accuracy of the model begins to gradually decrease with the increase of input features. Overall, the random forest model at each observation angle reaches the maximum at step 4 when RENDVI index is added. Therefore, the combination of GPRI, NPCI, VARI and RENDVI is determined as the optimal spectral feature combination as the input feature for the subsequent development of the corn southern rust monitoring model.

[0080] Example 8, Construction of Pest and Disease Detection Model

[0081] Based on the determined optimal spectral feature combination at each observation angle (GPRI, NPCI, VARI and RENDVI), a corn southern rust monitoring model was constructed using the random forest algorithm to evaluate the potential of spectral observation data at different angles in corn southern rust monitoring. The random forest algorithm uses multiple decision trees to make predictions, has the advantages of sample randomization and variable randomization, and can effectively avoid the influence of multicollinearity, and finally establishes a model for detecting corn southern rust.

[0082] Example 9, Plant Pest and Disease Occurrence Evaluation

[0083] The sample to be detected is placed in the model constructed in Example 8, and the multi-angle spectral enhanced specificity features of the corn sample to be detected are evaluated to determine the occurrence of plant pests and diseases. The accuracy of the evaluation and determination is evaluated using the overall accuracy (Formula 1) and F1 score (Formula 2) as evaluation indexes.

[0084]

[0085] In the formula, TP represents true positive; FP represents false positive; TN represents true negative; and FN represents false negative. ​

[0086] Example 10, Model accuracy evaluation comparison

[0087] Approximately 80% of the data was randomly selected as the training sample, and the remaining approximately 20% was used as the invisible data during the model training process (i.e., the validation sample) to evaluate the accuracy of the established model (Table 2). The model based on random forest is approximately parameter-free, and does not require parameter adjustment and optimization.

[0088] Table 2: Number of samples for model training and validation

[0089]

[0090] Example 11, Disease and pest monitoring potential evaluation

[0091] The sample to be detected was put into the model constructed in Example 8 to evaluate the potential of multi-view remote sensing in corn southern rust monitoring.

[0092] Figure 4 The overall accuracy and F1 score of different corn southern rust monitoring models.

[0093] The results showed that the overall accuracy of the 0° observation angle monitoring model was 64.2%, and the F1 score was 0.64. The overall accuracy of the inclined angle monitoring model was higher than that of the 0° observation angle monitoring model in corn southern rust monitoring. Among them, the accuracy of the 15° observation angle was the highest, with an overall accuracy of 70.8% and an F1 score of 0.71, which was 6.6% higher than that of the 0° observation angle and about 0.07 higher in F1 score; the accuracy of the 30° observation angle was second, with an overall accuracy of 69.6% and an F1 score of 0.70, which was 5.4% higher than that of the 0° observation angle and about 0.06 higher in F1 score.

[0094] The multi-view monitoring model simultaneously added spectral features under 0°, 15°, and 30° observation angles when the model was established, which could collect complementary spectral information of different layers of crops and effectively overcome the canopy shielding. The results showed that it significantly improved the accuracy of corn southern rust monitoring, and the final detection accuracy reached 73.7%, which was about 10% higher than that based on the spectral features of the 0° observation angle in overall accuracy and F1 score, indicating that the model had good detection accuracy.

[0095] Example 12, Drawing field disease and pest intensity distribution map

[0096] Based on the optimal model (multi-view observation model), the distribution map of corn southern rust disease and pest intensity in the field was drawn. Figure 5For corn southern rust monitoring mapping, the left image is a high-definition image of the field, and the right image is the intensity of corn southern rust based on the multi-view optimal model. The disease and pest intensity drawn in the figure has high consistency with the color change of the high-definition RGB corn canopy.

[0097] In addition, according to the mapping results, corresponding control measures can be taken for different disease and pest intensities in agricultural production management, and the spread and spread of diseases and pests can be controlled with emphasis, which is helpful for agricultural precision management and effectively reduces the use of pesticides.

Claims

1. A method for constructing a crop disease and pest monitoring model, comprising the following steps: 1) collecting unmanned aerial vehicle multi-view multi-spectral image data of a crop canopy, wherein the unmanned aerial vehicle multi-view multi-spectral image data is unmanned aerial vehicle multi-view multi-spectral image data collected at a wavelength range of 350-900 nm at an observation angle of 0-90° of the crop canopy; 2) respectively pre-processing the collected unmanned aerial vehicle multi-view multi-spectral image data, including image stitching, geometric correction, radiation correction, and canopy extraction; 3) using a random forest algorithm to extract a crop area from the unmanned aerial vehicle multi-view multi-spectral image data pre-processed in step 2), classifying bare soil and weeds in the image as non-crop areas, and obtaining a binary classification mask of crops / non-crops, which is used for extraction and calculation of disease and pest sensitive spectral indices; 4) based on wavebands sensitive to crop diseases and pests, using different combinations of generations to calculate spectral indices sensitive to diseases and pests for the crop area extracted in step 3); 5) using a random forest algorithm to evaluate the importance of each spectral index at different observation angles in disease and pest monitoring in step 4) by using an average decrease in Gini coefficient; 6) using a forward selection algorithm to determine the optimal combination of spectral indices at different observation angles; 7) using a random forest algorithm to construct a crop disease and pest monitoring model.

2. The method of claim 1, wherein: In the unmanned aerial vehicle multi-view image data collection of step 1), the unmanned aerial vehicle multi-view multi-spectral image data is unmanned aerial vehicle multi-view multi-spectral image data collected at an observation angle of 0-30° of the crop canopy. 3.The method of claim 1, wherein: In the unmanned aerial vehicle multi-view image data collection of step 1), the unmanned aerial vehicle multi-view multi-spectral image data is unmanned aerial vehicle multi-view multi-spectral image data collected at a wavelength range of 400-870 nm of the crop canopy.

4. The method of claim 2, wherein the method further comprises: determining a disease or pest occurrence probability of the crop based on the determined disease or pest occurrence probability of the crop and the determined disease or pest occurrence probability of the neighboring crop. In the unmanned aerial vehicle multi-view image data collection of step 1), the unmanned aerial vehicle multi-view multi-spectral image data is unmanned aerial vehicle multi-view multi-spectral image data collected at a wavelength range of 400-870 nm of the crop canopy. 5.The method of claim 1, wherein: In step 4), the spectral indices sensitive to diseases and pests are spectral indices closely related to various physiological and biochemical characteristics of crops.

6. The method of claim 2, wherein: In step 4), the spectral indices sensitive to diseases and pests are spectral indices closely related to various physiological and biochemical characteristics of crops.

7. The method of claim 3, wherein the method further comprises: receiving a request for a crop disease and pest monitoring model; and providing the crop disease and pest monitoring model to the user device. In step 4), the spectral indices sensitive to diseases and pests are spectral indices closely related to various physiological and biochemical characteristics of crops. 8.The method of claim 4, wherein: In step 4), the spectral indices sensitive to diseases and pests are spectral indices closely related to various physiological and biochemical characteristics of crops. 9.The method of claim 1, wherein: In step 5), the average decrease in Gini coefficient is used to quantify the influence of the heterogeneity information observed by the input feature in the decision tree node, so as to reflect the importance of the input feature.

10. The method of claim 2, wherein: In step 5), the average decrease in Gini coefficient is used to quantify the influence of the heterogeneity information observed by the input feature in the decision tree node, so as to reflect the importance of the input feature.

11. The method of claim 3, wherein: In step 5), the average decrease in Gini coefficient is used to quantify the influence of the heterogeneity information observed by the input feature in the decision tree node, so as to reflect the importance of the input feature.

12. The method of claim 4, wherein: In step 5), the average decrease in Gini coefficient is used to quantify the influence of the heterogeneity of the observations on the node of the decision tree, so as to reflect the importance of the input feature.

13. The method of claim 5, wherein: In step 5), the average decrease in Gini coefficient is used to quantify the influence of the heterogeneity of the observations on the node of the decision tree, so as to reflect the importance of the input feature.

14. The method of claim 6, wherein: In step 5), the average decrease in Gini coefficient is used to quantify the influence of the heterogeneity of the observations on the node of the decision tree, so as to reflect the importance of the input feature.

15. The method of claim 7, wherein: In step 5), the average decrease in Gini coefficient is used to quantify the influence of the heterogeneity of the observations on the node of the decision tree, so as to reflect the importance of the input feature.

16. The method of claim 8, wherein: In step 5), the average decrease in Gini coefficient is used to quantify the influence of the heterogeneity of the observations on the node of the decision tree, so as to reflect the importance of the input feature.

17. The method of claim 1 to 16, wherein: In step 6), the input spectral index includes spectral indices of different viewing angles and spectral indices of fusion of multiple viewing angles.

18. The method of claim 1 to 16, wherein: In step 6), the forward selection algorithm starts from an empty model, and iteratively adds spectral indices to the model according to the importance ranking, so that the model performance is maximally improved compared with the previous step.

19. The method of claim 17, wherein: In step 6), the forward selection algorithm starts from an empty model, and iteratively adds spectral indices to the model according to the importance ranking, so that the model performance is maximally improved compared with the previous step.

20. The method of claim 18, wherein: In step 6), the performance of each step of model construction is evaluated according to the accuracy of the random forest model, and the forward selection process continues to add the most important spectral indices one by one until all indices are considered.

21. The method of claim 19, wherein: In step 6), the performance of each step of model construction is evaluated according to the accuracy of the random forest model, and the forward selection process continues to add the most important spectral indices one by one until all indices are considered.

22. The method of claim 1 to 16, wherein: The crop disease and pest is corn southern rust.

23. The method of claim 17, wherein: The crop disease and pest is corn southern rust.

24. The method of claim 18, wherein: The crop disease and pest is corn southern rust.

25. The method of claim 19, wherein: The crop disease and pest is corn southern rust.

26. The method of claim 20, wherein: The crop disease and pest is corn southern rust.

27. The method of claim 21, wherein: The crop disease and pest is corn southern rust.

28. Use of the method for constructing the crop disease and pest monitoring model according to any one of claims 1 to 27 in crop disease and pest monitoring or crop disease and pest control.

Images