Farmland pest monitoring method and device and storage medium

Abstract

The application provides a farmland disease and pest monitoring method and device and a storage medium, relates to the technical field of agriculture, and divides the farmland into multiple sub-regions, acquires an infrared spectrum image and divides the infrared spectrum image into corresponding sub-images; based on the reflectivity of a pest-related characteristic wave band, a vegetation stress comprehensive index is calculated, and a high-risk sub-region is screened. In the high-risk region, risk plants are randomly selected and leaf physiological parameters are collected, a leaf physiological health index is constructed, and dynamic correction is performed according to environmental data through an environmental regulation coefficient. A trained network model is used to identify risk sample leaf lesions, and a lesion area index is calculated. Finally, the corrected leaf physiological health index and the lesion area index are fused, a normalized risk value is calculated through a risk assessment function, and four risk levels are divided. Through multi-level fusion of spectrum analysis, physiological parameter correction and deep learning lesion detection, the application realizes early and accurate warning of diseases and pests and improves monitoring efficiency.

Description

Technical Field

[0001] The present invention relates to the field of agricultural technology, and in particular to a method, device and storage medium for monitoring farmland pests and diseases. Background Art

[0002] With the development of modern agriculture, farmland pest and disease monitoring technology is playing an increasingly important role in ensuring food security and improving agricultural production efficiency. Currently, farmland pest and disease monitoring primarily relies on manual inspections, ground-based sensor monitoring, and remote sensing technology. Traditional farmland pest and disease monitoring relies primarily on agricultural technicians regularly patrolling farmland and visually observing crop growth and pest and disease symptoms. While intuitive, this method is inefficient, difficult to cover large areas of farmland, and susceptible to human influence, resulting in subjective and inaccurate monitoring results. With the advancement of science and technology, remote sensing technology is increasingly being used in agriculture. In particular, methods using drones equipped with hyperspectral cameras to acquire farmland information have become a research hotspot. Prior art publication CN112634212A discloses a method and system for detecting disease-prone trees using a hyperspectral drone. This method regularly collects leaf spectral data, combines it with satellite remote sensing imagery to determine disease-infested areas, and then establishes a disease data model to achieve early detection of diseased plants. Although this method has achieved certain results in forest pest and disease monitoring, its model does not consider the interference of environmental factors on spectral data and lacks analysis of microscopic lesion characteristics.

[0003] Existing farmland pest and disease monitoring technologies face the following major challenges: First, traditional single-use monitoring methods struggle to achieve early identification of pests and diseases. Most methods can only effectively detect pests and diseases after symptoms become apparent, resulting in delayed prevention and control measures and missed opportunities for optimal control. Second, environmental factors such as temperature, humidity, and light interfere with monitoring data, and existing technologies lack effective correction mechanisms for these environmental factors, which can easily lead to misinterpretations of monitoring results. Third, existing technologies often focus on data collection and analysis in a specific area, lacking comprehensive assessment methods that effectively integrate macroscopic spectral information, microscopic physiological parameters, and visual features of lesions, making it difficult to comprehensively and accurately assess pest and disease risks. Finally, most risk assessment models use fixed risk thresholds, which are unable to adapt to the basic conditions and environmental conditions of different farmlands, resulting in regional assessment bias and compromising the scientificity and effectiveness of prevention and control decisions. Although these methods integrate multi-source data, they lack monitoring of crop microphysiological parameters, making it difficult to achieve early warning of pests and diseases.

[0004] The above information disclosed in this Background section is only for enhancement of understanding of the background of the present disclosure and therefore it may contain information that does not form the prior art that is already known to a person of ordinary skill in the art. Summary of the Invention

[0005] The purpose of the present invention is to provide a method, device and storage medium for monitoring farmland pests and diseases, so as to solve the problems raised in the above background technology.

[0006] To achieve the above object, the present invention provides the following technical solutions:

[0007] The specific steps of the farmland pest and disease monitoring method include:

[0008] S1: Divide the farmland to be monitored into multiple sub-areas, obtain infrared spectral images of the farmland to be monitored, and divide the infrared spectral images into sub-images mapped to each sub-area. Calculate the vegetation stress comprehensive index based on the reflectance of the pest-related characteristic bands of the pixels in each sub-image, analyze and screen out high-risk sub-images, and screen out high-risk farmland areas within the sub-areas based on the mapping relationship;

[0009] S2: Select crop plants from each high-risk area as risk samples, measure the leaf physiological parameters of each risk sample, construct a leaf physiological health index, and correct the leaf physiological health index based on the environmental data when the risk sample was collected;

[0010] S3: Capture images of designated risk sample leaves and use the trained network model to identify lesions on the risk sample images; the model outputs leaf lesion data and evaluates the leaf lesion condition;

[0011] S4: Analyze the leaf physiological health index and lesion area index corrected by the risk samples corresponding to the high-risk area to obtain the pest and disease risk assessment index corresponding to the high-risk area; obtain the risk value through the risk assessment function, and classify the risk level according to the risk value.

[0012] Furthermore, the farmland was divided into 10m×10m grid sub-areas. An infrared spectral image of each sub-area of ​​the monitored farmland was obtained using a hyperspectral camera mounted on an unmanned aerial vehicle (UAV). The infrared spectral image was then segmented into sub-images corresponding to each sub-area. A multi-dimensional spectral feature analysis was performed on each pixel in the sub-image. The reflectivity was calculated for all pixels in the near-infrared band, red light band, and sub-area i in the 750nm and 700nm bands. The comprehensive vegetation stress index of each sub-image was calculated based on the reflectivity of the characteristic values ​​of each band. The formula is as follows:

[0013]

[0014] Among them, S i is the comprehensive index of vegetation stress, NDVI is the normalized difference vegetation index, i is the normalized vegetation index of sub-region i, R NIR is the average reflectivity of all pixels in region i in the near-infrared band, RRed is the average reflectance of all pixels in sub-region i in the red light band, NDVI h The health benchmark value, is the ratio of the average reflectance of all pixels in sub-region i at 750nm and 700nm, that is, the chlorophyll sensitive band ratio, The maximum ratio of the chlorophyll sensitive band in the whole field is set, and the threshold is set. i When it is >0.35, the sub-region is marked as a high-risk area.

[0015] Furthermore, randomly sample crops in high-risk sub-areas and use a fluorescence-stomatal meter to contact fully expanded leaves at the top of the plants to measure leaf physiological parameters, including stomatal conductance, transpiration rate, and photosynthetic rate. The maximum photosynthetic rate, optimal stomatal conductance, and maximum stomatal conductance of the crop varieties currently planted in the farmland are read. The physiological health of the leaves is evaluated and a leaf physiological health index is constructed using the following formula:

[0016]

[0017] Among them, Gs is the stomatal conductance, Gs max is the maximum stomatal conductance, Gs opt is the optimum stomatal conductance, E is the transpiration rate, A is the photosynthetic rate, T max is the maximum photosynthetic rate.

[0018] Furthermore, the leaf physiological health index is corrected based on the environmental data when the risk sample is collected. The environmental data when the risk sample is collected include: reading the temperature and humidity through the temperature sensor; reading the light intensity through the photon sensor; calling the crop parameter library to read the optimal humidity, optimal temperature, optimal light, and light saturation point of the crop variety currently planted in the farmland; and calculating the environmental control coefficient. The formula is as follows:

[0019]

[0020] Where I is temperature, H is humidity, I is light intensity, T opt is the optimum temperature, H opt The optimum humidity is I opt The optimal light intensity is obtained by calling the real-time environmental data of the risk sample location to correct the assessed physiological health status of the leaves and modifying the SPI value according to the following formula:

[0021] SPI′=SPI×(1-η×g(T,H,I))

[0022] Where η is the crop environmental sensitivity coefficient.

[0023] Furthermore, the trained network model was used to identify disease spots on the images of risk samples. The network model is an improved YOLOv11 network model, which uses CSPDarknet53 as the backbone network and combines it with PANet for feature fusion. It introduces an adaptive anchor point mechanism, which can automatically adjust the anchor point size according to the characteristics of the dataset, perform multi-channel decomposition on the input image, and detect real-time disease spots and insect-infested areas on the leaves in the image. The collected image is input, and the model outputs the disease spot area and the leaf imaging area. Density calculation is performed to obtain the disease spot area index, which is calculated as follows:

[0024]

[0025] Among them, A k is the area of ​​the kth lesion, A leaf is the leaf imaging area, CI is the image clarity index, and the lesion area index is output for pest and disease risk assessment.

[0026] Furthermore, the improved YOLOv11 network model obtains multiple farmland crop images, performs pathological feature enhancement preprocessing on the images, and marks the pest and disease areas to construct a training data set. The YOLOv11 network model is built, and the backbone network is connected in sequence by an initial convolution block, four groups of convolution blocks, a C3k2-Dual module, a multi-scale attention mechanism module, and an ASPP module; the neck network consists of a cross-scale connection structure with four layers of feature fusion; the detection head network includes three detection heads connected to 160×160, 40×40, and 10×10 scale features, respectively; the core formula of the feature fusion operation is as follows:

[0027] F out =Conv 1×1 (Concat[F s ,DenseASPP(F m ),ECA-Net(F l )])

[0028] Among them, Conv 1×1 is a 1×1 convolution operation, F s is the input shallow feature map, F m is the input middle-level feature map, F lFor input deep feature map, Concat(·) tensor splicing module, DenseASPP(·) is a dense atrous spatial pyramid pooling module, ECA-Net(·) is an efficient channel attention module, the crop image in the farmland is taken as a training set, the model is trained, the image in the test set is substituted into the model, and the corresponding detection result is obtained;The error between the detection result and the actual value in the test set is calculated;Determine whether the error meets the preset error threshold;If it meets, output the trained model, that is, the trained YOLOv11 network model;If it does not meet, return to continue training;The error is the mean absolute error, root mean square error and determination coefficient of the prediction result and the actual value in the test set.

[0029] Further, a disease and pest risk assessment function is constructed, a risk value is obtained through the risk assessment function, normalization processing is performed, and the risk level is classified according to the dynamic risk value;The disease and pest risk assessment function is as follows:

[0030]

[0031] Wherein, R risk is a disease and pest risk assessment index, SPI max is a maximum leaf physiological health index, and alpha and beta are weight coefficients, and alpha and beta are greater than zero, and alpha+beta=1;R risk is normalized to obtain R morm .

[0032] Further, a risk value is obtained through the risk assessment function, and the risk level is classified according to the risk value, and the classification rules are as follows: R norm <0.3, grade I, and the sub-area has no obvious stress;0.3≤R norm <0.6, grade II, the sub-area has early latent disease and pests;0.6≤R norm <0.8, grade III, the sub-area is in the rapid spread period of disease spots;R norm ≥0.8, grade IV, the sub-area may reduce production by >15%.

[0033] The application further provides a farmland disease and pest monitoring device, which is used for executing the farmland disease and pest monitoring method, and comprises:

[0034] A high-risk area screening module: the farmland to be monitored is divided into a plurality of sub-areas, an infrared spectrum image of the farmland to be monitored is obtained, and the infrared spectrum image is divided into a sub-image corresponding to each sub-area, a vegetation stress comprehensive index is calculated according to the reflectivity of the pixel disease and pest related characteristic band in each sub-image, and a high-risk sub-image is screened out by analysis, and a farmland high-risk area is screened out in the sub-area according to the mapping relationship;

[0035] Leaf physiological detection module: select crop plants in each high-risk area as risk samples, measure the leaf physiological parameters of each risk sample, construct a leaf physiological health index, and correct the leaf physiological health index according to the environmental data collected when the risk sample is collected;

[0036] Lesion detection module: capture the image of the specified risk sample leaf, and use the trained network model to identify the lesion of the risk sample image; the leaf lesion data output by the model is used to calculate the lesion area index;

[0037] Disease and pest risk assessment and grading module: analyze the corrected leaf physiological health index and lesion area index of the risk sample corresponding to the high-risk area to obtain the disease and pest risk assessment index corresponding to the high-risk area; obtain the risk value through the risk assessment function, and grade the risk level according to the risk value.

[0038] The application further provides a storage medium, characterized in that the storage medium stores a computer program, and when the computer program is executed by a processor, the computer program is used to implement the farmland disease and pest monitoring method.

[0039] Compared with the prior art, the application has the beneficial effects that: through the step-by-step verification mechanism of spectral macroscopic screening, physiological microscopic measurement and lesion visual positioning, the effective fusion of multi-source data is realized, the accuracy and reliability of disease and pest monitoring are significantly improved, and the identification ability in the early stage of disease and pest is obviously enhanced; the SPI obtained by collecting leaf physiological parameters can indirectly reflect the phenomenon of disease and pest of crop roots, so that the disease and pest detection is not limited to judging the disease and pest risk based on the surface characteristics of the leaf, and the SPI can also indirectly reflect the root disease and pest, which is conducive to early detection of latent disease and pest; through the leaf health index, the interference of meteorological factors such as temperature, humidity and illumination is effectively eliminated by using the SPI correction model, the monitoring misjudgment caused by environmental changes is avoided, the system can work stably under different meteorological conditions, the grid partition design matches the unmanned aerial vehicle aerial photography specification, the YOLOv11 model is lightweight and suitable for embedded devices, so that the system is easy to deploy and implement, and the efficiency is improved compared with the traditional manual inspection; the normalization processing based on real-time data extreme value avoids the regional evaluation deviation caused by the fixed threshold, so that the risk assessment result is more objective and accurate, and is suitable for different farmland basic states, and in the application test in different regions, the risk assessment accuracy is more accurate than that of the traditional method. BRIEF DESCRIPTION OF DRAWINGS

[0040] Figure 1 It is a whole method flowchart of the application;

[0041] Figure 2 It is an infrared spectrum and leaf physiological parameter data graph of the application embodiment experiment.

[0042] Figure 3 Normalized risk index data graph for the embodiment experiment of the present application;

[0043] Figure 4 Device module schematic diagram of the present application. DETAILED DESCRIPTION

[0044] In order to make the objectives, technical solutions and advantages of the present application clearer, the present application will be further described in detail below with specific embodiments.

[0045] It should be noted that, unless otherwise defined, the technical terms or scientific terms used in the present application should be understood as the general meaning understood by those skilled in the art to which the present application belongs. The "first", "second" and similar words used in the present application do not represent any order, quantity or importance, but are only used to distinguish different components. "Include" or "contain" and similar words mean that the elements or objects before the word cover the elements or objects listed after the word and their equivalents, and do not exclude other elements or objects. "Connected" or "connected" and similar words are not limited to physical or mechanical connection, but can include electrical connection, whether direct or indirect. "Up", "down", "left", "right" and the like are only used to represent the relative positional relationship, when the absolute position of the described object changes, the relative positional relationship may also change accordingly.

[0046] Embodiment:

[0047] Please refer to Figures 1 to 3 The present application provides a technical solution:

[0048] The method for monitoring the crop disease and insect pest, the specific steps include:

[0049] S1: dividing the farmland to be monitored into multiple sub-regions, acquiring the infrared spectrum image of the farmland to be monitored, dividing the infrared spectrum image into sub-images mapped with each sub-region, calculating the vegetation stress comprehensive index according to the reflectivity of the pixel disease and insect pest related characteristic band in each sub-image, and analyzing and screening out the high-risk sub-image, screening out the high-risk area of the farmland in the sub-region according to the mapping relationship;

[0050] In this embodiment, the farmland is divided into 10m*10m grid sub-regions, and a hyperspectral camera is carried by a UAV to obtain infrared spectral images of each sub-region of the farmland to be monitored as sub-images, and the infrared spectral images are simultaneously segmented into sub-images corresponding to each sub-region; multi-dimensional spectral feature analysis is performed on each pixel point of the sub-images, and the reflectivity is calculated according to the average value of all pixels in the near-infrared band (760-900nm band), the average value of all pixels in the red light band (630-690nm band), and the reflection of the sub-region i in the 750nm and 700nm bands; according to the reflectivity of the characteristic values of each band, the vegetation stress comprehensive index of each sub-image is calculated, and the formula is as follows:

[0051]

[0052] wherein, S i is the vegetation stress comprehensive index, NDVI is the normalized vegetation index, NDVI i is the normalized vegetation index of the sub-region i, R NIR is the average value of the reflectivity of all pixels in the near-infrared band in the region i, R Red is the average value of the reflectivity of all pixels in the red light band in the sub-region i, NDVI h is the healthy baseline value, is the average value of the reflectivity of all pixels in the sub-region i in the 750nm and 700nm bands, i.e. the chlorophyll sensitive band ratio, is the maximum ratio of the chlorophyll sensitive band ratio of the whole field; the biophysical meaning of NDVI is that the healthy leaf sponge tissue strongly reflects the NIR band, the chlorophyll absorbs the Red band, the disease and insect pests cause the destruction of the leaf structure, the NIR band reflectivity decreases, the chlorophyll explains the Red band reflectivity increases, which leads to the decrease of the NDVI value, and vice versa, the healthy vegetation has a higher NDVI value; when S i is calculated, the molecular part "NDVI h -NDVI i characterizes the degree of growth attenuation of the current sub-region relative to the healthy vegetation; a threshold is set, when S i >0.35, the sub-region is marked as a high-risk area. The threshold 0.35 has a clear biological meaning: if the NDVI of the sub-region decreases to 70% of the healthy value, the chlorophyll activity decreases to 10% of the highest value in the whole field; at this time, S i= 0.3 + 0.05 = 0.35; the threshold value can capture both growth recession and physiological disorder signals, reducing the false negative rate; the identification of high-risk areas is essentially through the inversion of vegetation physiological state by infrared spectral characteristics: NDVI decline indicates photosynthetic tissue damage, R750 / 700 abnormality indicates chlorophyll metabolism disorder, and the coupling of the two indicators helps to exclude single environmental interference; the screening of high-risk areas is realized through the vegetation physiological stress index inverted by infrared spectrum, which integrates the dual signals of photosynthetic capacity decline and chlorophyll metabolism abnormality, and the threshold setting takes into account both sensitivity and economy.

[0053] This embodiment selects three test groups: A: healthy control group, B: disease incubation group, C: pest outbreak group; at four time points: 1st day, 5th day, 10th day, 15th day, experimental data measurement is carried out, and comparison is made with traditional measurement method warning, the environment involved in A group is healthy environment, and any sub-area data is randomly collected, the experimental data of B and C groups are S i > 0.35 high-risk sub-area data, this experiment is used to show the advantages of this technical solution compared with traditional method in disease and pest monitoring; the experimental measurement data are as follows:

[0054] Table 1: NDVI and R750 / R700 measurement data table of test groups

[0055] Monitoring schedule 1 1 1 5 5 5 10 10 10 15 15 15 experimental group A B C A B C A B C A B C NDVI 0.78 0.72 0.65 0.79 0.68 0.59 0.77 0.63 0.51 0.76 0.57 0.42 R750 / R700 1.28 1.18 1.05 1.3 1.12 0.97 1.29 1.05 0.88 1.27 0.95 0.79

[0056] S2: select crop plants in each high-risk area as risk samples, measure the leaf physiological parameters of each risk sample, construct a leaf physiological health index, and correct the leaf physiological health index according to the environmental data collected when the risk sample is collected;

[0057] In this embodiment, the sample crops in the high-risk sub-area are randomly inspected, the fluorescence-stomatal measuring instrument is used to contact the fully expanded leaves at the top of the plant, the top leaves are sensitive to stress, contact measurement can avoid environmental interference and ensure data reliability, leaf physiological parameters are measured, and the leaf physiological parameters include: stomatal conductance, transpiration rate and photosynthetic rate; the maximum photosynthetic rate, the optimum stomatal conductance and the maximum stomatal conductance of the crop variety planted in the current farmland are read from the preset library; the leaf physiological health state is evaluated, and a leaf physiological health index is constructed, and the formula is as follows:

[0058]

[0059] Gs is the stomatal conductance, Gs max is the maximum stomatal conductance, Gs opt is the optimum stomatal conductance, E is the transpiration rate, A is the photosynthetic rate, and A max is the maximum photosynthetic rate. It represents the photosynthetic efficiency ratio. The closer the value is to 1, the healthier the environment is. The lower the value is, the more severe the stress is. exp(-|·|) represents the inconsistency between stomatal conductance Gs and photosynthetic rate A. The larger the absolute value is, the more severe the decay of the exponential term is. Carbon-water balance index, the larger the value, the higher the water use efficiency; It is the loss caused by stomatal aperture deviating from the optimal value; the composite index comprehensively reflects photosynthetic efficiency, stomatal regulation and water use efficiency, and is more sensitive than a single parameter.

[0060] The leaf physiological health index is corrected based on the environmental data when the risk sample is collected. The environmental data when the risk sample is collected include: reading the temperature and humidity through the temperature sensor; reading the light intensity through the photon sensor; calling the crop parameter library to read the optimal humidity, optimal temperature, optimal light, and light saturation point of the crop variety currently planted in the farmland; and calculating the environmental control coefficient. The formula is as follows:

[0061]

[0062] Where T is temperature, H is humidity, I is light intensity, T opt is the optimum temperature, H opt The optimum humidity is I opt is the optimal light intensity; High humidity penalty, only when H>H opt The denominator is limited to 30 to avoid over-correction; The exponential decay of temperature causes T to deviate from T opt The coefficient decreases by 50% at ±5℃; It is the light deviation penalty, linear penalty for insufficient light (I opt ) or strong light stress (I>I opt ), call the real-time environmental data of the risk sample location, correct the assessed leaf physiological health status, and modify the SPI value according to the following formula:

[0063] SPI′=SPI×(1-η×g(T,H,I))

[0064] Where η is the crop environmental sensitivity coefficient, eliminating environmental interference to ensure that the physiological index truly reflects the stress of pests and diseases. The experimental measurement data are as follows:

[0065] Table 2: Experimental data of SPI and SPI′ in the experimental group of the embodiment

[0066] Monitoring schedule 1 1 1 5 5 5 10 10 10 15 15 15 experimental group A B C A B C A B C A B C SPI 0.92 0.83 0.71 0.93 0.75 0.61 0.9 0.68 0.48 0.88 0.56 0.32 SPI′ 0.91 0.82 0.72 0.92 0.76 0.63 0.91 0.7 0.51 0.89 0.59 0.35

[0067] Reference Figure 2 , Figure 2 ​The figure clearly shows the leaf physiological parameters and environmental data of the three groups of farmland at four observation time points. The leaf physiological health index is calculated by collecting the results of leaf stomatal conductance, transpiration rate, and photosynthetic rate. The leaf physiological health index is then corrected for the environment using environmental data such as light intensity, humidity, and temperature.

[0068] S3: Capture images of designated risk sample leaves and use the trained network model to identify lesions on the risk sample images; the model outputs leaf lesion data and evaluates the leaf lesion condition;

[0069] In this embodiment, the trained network model is used to identify disease spots on the images of risk samples. The network model is an improved YOLOv11 network model, which uses CSPDarknet53 as the backbone network and combines it with PANet for feature fusion. It introduces an adaptive anchor point mechanism, which can automatically adjust the anchor point size according to the characteristics of the data set, perform multi-channel decomposition on the input image, and detect real-time disease spots and insect-infested areas in the image leaves. The collected image is input, and the model outputs the disease spot area and the leaf imaging area. The density calculation is performed to obtain the disease spot area index. The formula is as follows:

[0070]

[0071] Among them, A k is the area of ​​the kth lesion, A leaf is the leaf imaging area, CI is the image clarity index, when CI>0.7, the correction term is ≈1, when CI=0.5, the image is blurred, the correction term is 0.62, to avoid misjudgment of the lesion ratio, output the lesion area index, which is used for disease and insect pest risk assessment, improve the reliability of lesion quantification, and avoid risk misjudgment caused by image quality.

[0072] The improved YOLOv11 network model acquires multiple farmland crop images, performs pathological feature enhancement preprocessing on the images, and labels the pest and disease areas to construct a training dataset. The YOLOv11 network model is constructed, and the backbone network is sequentially connected by an initial convolution block, four groups of convolution blocks, a C3k2-Dual module, a multi-scale attention mechanism module, and an ASPP module. The neck network is a cross-scale connection structure with four layers of feature fusion. The detection head network includes three detection heads connected to 160×160, 40×40, and 10×10 scale features, respectively. The core formula of the feature fusion operation is as follows:

[0073] F out =Conv 1×1 (Concat[F s ,DenseASPP(F m ),ECA-Net(F l )])

[0074] Among them, Conv 1×1 is a 1×1 convolution operation, F s is the input shallow feature map, F m is the input middle-level feature map, F l The model uses a deep feature map as input, a tensor concatenation module called Concat(·), a dense atrous spatial pyramid pooling module called DenseASPP(·), and dense atrous convolution with dilation rates of 2, 4, 8, and 12 to capture lesions of varying sizes (0.5-5 mm). ECA-Net(·) is an efficient channel attention module. The model is trained using crop images as a training set. Images from the test set are substituted into the model to obtain corresponding detection results. The error between the detection results and the actual values ​​in the test set is calculated. A determination is made as to whether the error meets a preset error threshold. If so, the trained model, i.e., the trained YOLOv11 network model, is output. Otherwise, training continues. The error is the mean absolute error, root mean square error, and coefficient of determination between the predicted results and the actual values ​​in the test set. This improved model improves the recall rate of small lesions, addresses the variable size of leaf lesions, and provides accurate input for the lesion index.

[0075] S4: Analyze the leaf physiological health index and lesion area index corrected by the risk samples corresponding to the high-risk area to obtain the pest and disease risk assessment index corresponding to the high-risk area; obtain the risk value through the risk assessment function, and classify the risk level according to the risk value.

[0076] In this embodiment, a pest and disease risk assessment function is constructed, and a risk value is obtained through the risk assessment function, which is normalized and the risk level is graded according to the dynamic risk value. The pest and disease risk assessment function has the following formula:

[0077]

[0078] Among them, R risk SPI is the Pest and Disease Risk Assessment Index max is the maximum value of leaf physiological health index, α and β are weight coefficients; α = 0.6, β = 0.4, α> β emphasizes the intuitive harm of the lesion, D lesion Reflects the dominant characteristics of pests and diseases on leaves, early latent signals; SPI max Take the optimal value of the current farmland to avoid interference from variety differences; R risk Perform normalization processing;

[0079]

[0080] Among them, R min Minimum value of pest and disease risk assessment index, R maxIt is the minimum maximum value of the pest and disease risk assessment index; the normalized denominator is dynamic and normalized in real time.

[0081] In this embodiment, the risk value is obtained through the risk assessment function, and the risk level is graded according to the risk value. The grading rules are as follows: norm <0.3, level I, no obvious stress in the sub-area; 0.3≤R norm <0.6, level II, the sub-area has early-stage latent diseases and insect pests; 0.6≤R norm <0.8, Grade III, the lesion in this sub-area is rapidly spreading; R norm ≥0.8, Grade IV, the sub-area may experience a yield reduction of >15%; in this example, the pest risk assessment data for the three test groups are as follows in Table 3:

[0082] Table 3: Pest and disease risk assessment data for the experimental group

[0083] Monitoring schedule 1 1 1 5 5 5 10 10 10 15 15 15 experimental group A B C A B C A B C A B C Lesion area index 0.01 0.03 0.05 0.01 0.08 0.15 0.02 0.14 0.32 0.02 0.23 0.51 Risk Index 0.05 0.12 0.38 0.04 0.25 0.42 0.06 0.38 0.67 0.07 0.59 0.89 Normalized Value at Risk 0.02 0.1 0.3 0.01 0.22 0.39 0.03 0.35 0.64 0.04 0.56 0.86 Traditional methods of early warning 0 0 0 0 0 1 0 0 1 0 1 1

[0084] Reference Figure 3 Using an improved YOLOv11 network model, the trained outputs were the lesion area and leaf imaging area. The lesion area index was calculated based on the model's detection results. Finally, the leaf physiological health index and the lesion area index were combined to determine the risk index for the high-risk area. Finally, normalization was performed to unify the scale, eliminating the interference of inherent differences in soil fertility, variety characteristics, and growth period on risk assessment. The corresponding classification thresholds were used to detect farmland pest and disease risk areas, providing earlier pest and disease warnings than traditional methods. Group C issued a Level II warning on the first day, indicating an early potential for pests and diseases in this sub-area. Group B issued a Level II warning on the tenth day, indicating an early potential for pests and diseases in this sub-area. Both warnings were earlier than traditional methods. Furthermore, different risk levels facilitated agricultural workers' preparation for pest and disease prevention and treatment, tailored to the specific risk levels.

[0085] See also Figure 4 The present invention further provides a farmland pest and disease monitoring device: the farmland pest and disease monitoring device is used to perform the above-mentioned farmland pest and disease monitoring method, comprising:

[0086] High-risk area screening module: The farmland to be monitored is divided into multiple sub-areas, and infrared spectral images of the farmland to be monitored are obtained. The infrared spectral images are then divided into sub-images corresponding to each sub-area. Based on the reflectivity of the characteristic bands related to pests and diseases in each sub-image, the comprehensive vegetation stress index is calculated, and high-risk sub-images are screened out through analysis. Based on the mapping relationship, high-risk farmland areas are screened out within the sub-areas.

[0087] Leaf Physiology Detection Module: Crop plants in high-risk areas are randomly selected as risk samples. Leaf physiological parameters of each risk sample are measured to construct a leaf physiological health index. The leaf physiological health index is then corrected based on the environmental data when the risk samples were collected.

[0088] Disease spot detection module: Captures images of designated risk sample leaves and uses the trained network model to identify disease spots on the risk sample images; the model outputs leaf disease spot data and calculates the disease spot area index;

[0089] Pest and disease risk assessment and grading module: Analyze the leaf physiological health index and lesion area index corrected by the risk samples corresponding to the high-risk areas to obtain the pest and disease risk assessment index corresponding to the high-risk areas; obtain the risk value through the risk assessment function, and grade the risk level according to the risk value.

[0090] The present invention further provides a storage medium having a computer program stored thereon. When the computer program is executed by a processor, it is used to implement the above-mentioned method for monitoring farmland pests and diseases.

[0091] The above formulas are all dimensionless and numerical calculations. The formulas are obtained by collecting a large amount of data and performing software simulation to obtain the most recent real situation. The preset parameters in the formulas are set by technicians in this field according to actual conditions.

[0092] The above embodiments can be implemented in whole or in part by software, hardware, firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented in whole or in part in the form of a computer program product. Those skilled in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented by electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed by hardware or software depends on the specific application and design constraints of the technical solution.

[0093] The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, and may be located in one place or distributed across multiple network units. Some or all of these units may be selected to achieve the purpose of this embodiment as needed.

[0094] The above is only a specific implementation method of the present application, but the scope of protection of the present application is not limited thereto. Any technician familiar with this technical field can easily think of changes or replacements within the technical scope disclosed in this application, which should be covered by the scope of protection of the present application.

Claims

1. A method for monitoring farmland pests and diseases, characterized in that: The specific steps include: S1: Divide the farmland to be monitored into multiple sub-areas, obtain infrared spectral images of the farmland to be monitored, and divide the infrared spectral images into sub-images mapped to each sub-area. Calculate the vegetation stress comprehensive index based on the reflectance of the pest-related characteristic bands of the pixels in each sub-image, analyze and screen out high-risk sub-images, and screen out high-risk farmland areas within the sub-areas based on the mapping relationship; S2: Select crop plants from each high-risk area as risk samples, measure the leaf physiological parameters of the risk samples, and evaluate the leaf physiological health status through multi-parameter coupling. Based on the environmental data when the risk samples were collected, penalty items are designed based on the environmental data, and the assessed leaf physiological health status is corrected according to the environmental sensitivity of different crops. S3: Capture images of designated risk sample leaves and use the trained network model to identify lesions on the risk sample images. The model outputs the lesion area and leaf imaging area, and clarity compensation is performed based on the image clarity to evaluate the leaf lesion condition. S4: Analyze the leaf physiological health index and lesion area index corrected by the risk samples corresponding to the high-risk area to obtain the pest and disease risk assessment index corresponding to the high-risk area; obtain the risk value through the risk assessment function, and classify the risk level according to the risk value; Random sampling of sample crops in high-risk sub-areas was used. A fluorescence-stomatal meter was used to contact fully expanded leaves at the top of the plants to measure leaf physiological parameters, including stomatal conductance, transpiration rate, and photosynthetic rate. The maximum photosynthetic rate, optimal stomatal conductance, and maximum stomatal conductance of the crop varieties currently planted in the farmland were read. The physiological health of the leaves was assessed and a leaf physiological health index was constructed using the following formula: Among them, Gs is the stomatal conductance, Gs max is the maximum stomatal conductance, Gs opt is the optimum stomatal conductance, E is the transpiration rate, A is the photosynthetic rate, and A max is the maximum photosynthetic rate; The leaf physiological health index is corrected based on the environmental data when the risk sample is collected. The environmental data when the risk sample is collected include: reading the temperature and humidity through the temperature sensor; reading the light intensity through the photon sensor; calling the crop parameter library to read the optimal humidity, optimal temperature, optimal light, and light saturation point of the crop variety currently planted in the farmland; and calculating the environmental control coefficient. The formula is as follows: Where T is temperature, H is humidity, I is light intensity, T opt is the optimum temperature, H opt The optimum humidity is I opt The optimal light intensity is obtained by calling the real-time environmental data of the risk sample location to correct the assessed physiological health status of the leaves and modifying the SPI value according to the following formula: SPI′=SPI×(1-η×g(T,H,I)) Wherein, η is the crop environmental sensitivity coefficient; The trained network model is used to identify disease spots on images of risk samples. The network model is an improved YOLOv11 network model, which uses CSPDarknet53 as the backbone network and combines it with PANet for feature fusion. It introduces an adaptive anchor point mechanism, which can automatically adjust the anchor point size according to the characteristics of the dataset. It performs multi-channel decomposition on the input image and detects real-time disease spots and insect-infested areas on the leaves in the image. The collected image is input, and the model outputs the disease spot area and the leaf imaging area. The density calculation is performed to obtain the disease spot area index. The formula is as follows: Among them, A k is the area of ​​the kth lesion, A leaf is the leaf imaging area, CI is the image clarity index, and the lesion area index is output for pest and disease risk assessment.

2. The method for monitoring farmland pests and diseases according to claim 1, wherein: The farmland was divided into 10m×10m grid sub-areas. An infrared spectral image of each sub-area of ​​the monitored farmland was obtained using a hyperspectral camera mounted on an unmanned aerial vehicle (UAV). The infrared spectral image was then segmented into sub-images corresponding to each sub-area. A multi-dimensional spectral feature analysis was performed on each pixel in the sub-image. The reflectance was calculated for all pixels in the near-infrared band, red band, and sub-area i in the 750nm and 700nm bands. The comprehensive vegetation stress index of each sub-image was calculated based on the reflectance of the characteristic values ​​of each band. The formula is as follows: Among them, S i is the comprehensive index of vegetation stress, NDVI is the normalized difference vegetation index, i is the normalized vegetation index of sub-region i, R NIR is the average reflectivity of all pixels in region i in the near-infrared band, R Red is the average reflectance of all pixels in sub-region i in the red light band, NDVI h The health benchmark value, is the ratio of the average reflectance of all pixels in sub-region i at 750nm and 700nm, that is, the chlorophyll sensitive band ratio, The maximum ratio of the chlorophyll sensitive band in the whole field is set, and the threshold is set. i When it is >0.35, the sub-region is marked as a high-risk area.

3. The method for monitoring farmland pests and diseases according to claim 1, wherein: The improved YOLOv11 network model acquires multiple farmland crop images, performs pathological feature enhancement preprocessing on the images, and labels pest and disease areas to construct a training dataset. A YOLOv11 network model is constructed, wherein the backbone network is sequentially connected by an initial convolutional block, four groups of convolutional blocks, a C3k2-Dual module, a multi-scale attention mechanism module, and an ASPP module. The neck network is a cross-scale connection structure with four layers of feature fusion. The detection head network includes three detection heads connected to 160×160, 40×40, and 10×10 scale features, respectively. The core formula of the feature fusion operation is as follows: F out =Conv 1×1 (Concat[F s ,DenseASPP(F m ),ECA-Net(F l )]) Among them, Conv 1×1 is a 1×1 convolution operation, F s is the input shallow feature map, F m is the input middle-level feature map, F l The deep feature map is input, Concat(·) is the tensor splicing module, DenseASPP(·) is the dense void spatial pyramid pooling module, ECA-Net(·) is the efficient channel attention module, and the farmland crop image is used as the training set to train the model. The images in the test set are substituted into the model to obtain the corresponding detection results; the error between the detection result and the actual value in the test set is calculated; it is determined whether the error meets the preset error threshold; if so, the trained model, that is, the trained YOLOv11 network model, is output; if not, the training is returned to continue; the error is the mean absolute error, root mean square error and determination coefficient between the prediction result and the actual value in the test set.

4. The method for monitoring farmland pests and diseases according to claim 1, wherein: Construct a pest and disease risk assessment function, obtain the risk value through the risk assessment function, perform normalization, and classify the risk level according to the dynamic risk value; the pest and disease risk assessment function has the following formula: Among them, R risk Pest and disease risk assessment index, SPI max is the maximum value of leaf physiological health index, α and β are weight coefficients, and both α and β are greater than zero, satisfying α + β = 1; R risk Normalization is performed to obtain R norm .

5. The method for monitoring farmland pests and diseases according to claim 4, wherein: The risk value is obtained through the risk assessment function, and the risk level is graded according to the risk value. The grading rules are as follows: R norm <0.3, level I, no obvious stress in this sub-area; 0.3≤R norm <0.6, level II, the sub-area has early-stage latent diseases and insect pests; 0.6≤R norm <0.8, Grade III, the lesion in this sub-area is rapidly spreading; R norm ≥0.8, Level IV, the sub-area may have a production reduction of >15%.

6. Farmland pest and disease monitoring device, characterized by: The farmland pest and disease monitoring device is used to perform the farmland pest and disease monitoring method according to any one of claims 1 to 5, comprising: High-risk area screening module: The farmland to be monitored is divided into multiple sub-areas, and infrared spectral images of the farmland to be monitored are obtained. The infrared spectral images are then divided into sub-images corresponding to each sub-area. Based on the reflectivity of the characteristic bands related to pests and diseases in each sub-image, the comprehensive vegetation stress index is calculated, and high-risk sub-images are screened out through analysis. Based on the mapping relationship, high-risk farmland areas are screened out within the sub-areas. Leaf Physiology Detection Module: Crop plants in high-risk areas are randomly selected as risk samples. Leaf physiological parameters of each risk sample are measured to construct a leaf physiological health index. The leaf physiological health index is then corrected based on the environmental data when the risk samples were collected. Disease spot detection module: Captures images of designated risk sample leaves and uses the trained network model to identify disease spots on the risk sample images; the model outputs leaf disease spot data and calculates the disease spot area index; Pest and disease risk assessment and grading module: Analyze the leaf physiological health index and lesion area index corrected by the risk samples corresponding to the high-risk areas to obtain the pest and disease risk assessment index corresponding to the high-risk areas; obtain the risk value through the risk assessment function, and grade the risk level according to the risk value.

7. A storage medium, characterized in that: The storage medium stores a computer program, which, when executed by a processor, is used to implement the farmland pest and disease monitoring method according to any one of claims 1 to 5.

Images