Cotton field disease and pest control pesticide spraying amount prediction system based on unmanned aerial vehicle multispectral remote sensing image

Abstract

The invention discloses a cotton field disease and pest control pesticide spraying amount prediction system based on an unmanned aerial vehicle multispectral remote sensing image, and belongs to the field of cotton field disease and pest control pesticide spraying amount prediction, and the system comprises a data acquisition module, an image alignment module, a panoramic spectral image generation module, and a disease and pest recognition and pesticide spraying amount prediction module. The data acquisition module controls the unmanned aerial vehicle to acquire a cotton field multispectral remote sensing image, records an image space-time position through a GPS and an IMU, and collects image processing modes in different historical weathers; the image alignment module aligns the image with the shooting time, divides small areas of the cotton field and associates real-time weather; the panoramic spectral image generation module screens high-quality images and splices the high-quality images into a cotton field panoramic image; the disease and insect pest recognition module recognizes disease and insect pests through deep learning, quantifies the stress degree and generates a variable pesticide application prescription map; the system can improve image quality and processing efficiency, realizes accurate pesticide spraying amount prediction, reduces pesticide waste, and effectively solves the problem of low pesticide application accuracy in traditional cotton field pest and disease control.

Description

Technical Field

[0001] This invention belongs to the field of predicting the amount of pesticides sprayed for the control of cotton pests and diseases, specifically a system for predicting the amount of pesticides sprayed for the control of cotton pests and diseases based on UAV multispectral remote sensing images. Background Technology

[0002] In cotton field pest and disease control, precision spraying is key to improving control efficiency and reducing pesticide waste, which heavily relies on high-quality acquisition and efficient processing of multispectral remote sensing imagery from drones. However, existing technologies suffer from several pain points in practical applications: First, when drones capture images of cotton fields, the large area of ​​the fields often leads to long shooting times, during which the weather is prone to change. There is a lack of effective means to accurately correlate images with shooting time, specific cotton field areas, and corresponding real-time weather, resulting in inconsistent image-weather matching and affecting the accuracy of subsequent analysis. Second, there is no systematic collection of image processing methods under different weather conditions from historical data. When faced with images of sunny, cloudy, or other different weather conditions, processing methods must be explored on the spot, which is not only inefficient but also makes it difficult to ensure the consistency of image processing quality. These problems severely restrict the realization of precision control of cotton field pests and diseases, and there is an urgent need for a spray dosage prediction system that can solve the above pain points and provide reliable technical support for precision spraying. Summary of the Invention

[0003] This invention aims to at least solve one of the technical problems existing in the prior art. To this end, this invention proposes a cotton field pest and disease control spraying dosage prediction system based on UAV multispectral remote sensing imagery, enabling relevant personnel to conveniently predict the amount of pesticides to be sprayed for cotton field pests and diseases.

[0004] To achieve the above objectives, the present invention adopts the following technical solution:

[0005] A cotton field pest and disease control spraying dosage prediction system based on UAV multispectral remote sensing imagery includes a data acquisition module, an image alignment module, a panoramic spectral image generation module, and a pest and disease identification and spraying dosage prediction module.

[0006] The data acquisition module controls the drone to collect multispectral remote sensing images of cotton fields, and uses GPS and IMU sensors to record the spatiotemporal location of each image. At the same time, it collects historical data on the processing methods of images under different weather conditions.

[0007] The image alignment module aligns the images with the shooting time, divides the cotton field into multiple small areas based on the geographical information of the cotton field, and associates the real-time weather conditions of each area to determine the location of the cotton field and weather conditions corresponding to each image.

[0008] The panoramic spectral image generation module assesses the weather in each region, then uses the overlap of images taken by drones to remove images with poor weather conditions, and selects the images with the best weather conditions to stitch together a panoramic view of the cotton field. When there are no clear weather images for a certain cotton field area at any shooting time, the best version is directly selected from the available images and processed to make its visual and spectral characteristics resemble those of a clear weather image.

[0009] The pest and disease identification and pesticide application prediction module uses a deep learning model to identify areas in cotton fields infected with pests and diseases based on known panoramic multispectral images. By analyzing the spectral reflectance characteristics of specific bands, it quantifies and assesses the stress level of pests and diseases, generating variable pesticide application prescription maps for cotton fields.

[0010] Furthermore, the data acquisition module controls the drone to collect multispectral remote sensing images of the cotton field, and uses GPS and IMU sensors to record the spatiotemporal location of each image. Simultaneously, it collects historical data on image processing methods under different weather conditions. The processing steps are as follows:

[0011] Based on the geographical boundaries of the cotton fields and preset flight parameters including flight altitude, flight speed and overlap rate, a complete data acquisition route is automatically planned for the UAV. The UAV is then controlled to fly automatically along the predetermined route, and its onboard multispectral camera is instructed to take pictures at designated spatial points, thereby systematically collecting multispectral remote sensing images covering the entire cotton field area.

[0012] At the moment the drone flies and the camera captures the image, the onboard GPS and IMU sensors are simultaneously triggered to record the latitude and longitude coordinates of the shooting center and the attitude angles of the drone platform at that time, including roll, pitch, and yaw.

[0013] Historical data is collected from the system database, and processing methods for multispectral images under different weather conditions, including sunny and cloudy conditions, are collected.

[0014] All multispectral images, along with their corresponding data and historical data processing methods, are integrated into a structured dataset.

[0015] Furthermore, the image alignment module aligns the images with the shooting time, divides the cotton field into multiple small areas based on the geographical information of the cotton field, and associates the real-time weather conditions of each area to determine the location of the cotton field and weather conditions corresponding to each image. The processing procedure is as follows:

[0016] The data of each multispectral remote sensing image file is analyzed, and the precise capture timestamp and spatial coordinates of each image are extracted to establish the correspondence between the image and the capture time.

[0017] The system retrieves pre-entered geographical boundary information of the cotton field and divides the entire cotton field into multiple small areas according to the set rules. Then, based on the image capture time, it queries and associates the real-time weather conditions of the area at that moment with external meteorological data. The weather conditions include sunny and cloudy. Each image is then accurately assigned to a specific small area of ​​the cotton field and marked with the corresponding capture time and detailed real-time weather conditions.

[0018] Dividing the entire cotton field area into multiple small units or grids based on a geographic coordinate system enables refined spatial management, allowing each small area to be processed and analyzed independently.

[0019] Furthermore, the process involves calling pre-recorded geographical boundary information of the cotton field and dividing the entire cotton field into multiple small areas according to set rules. Then, based on the image capture time, it queries and associates the real-time weather conditions matching the region at that moment from external meteorological data. The weather conditions include sunny and cloudy. Each image is then precisely assigned to a specific small cotton field area and marked with its corresponding capture time. The detailed real-time weather processing process is as follows:

[0020] The system retrieves pre-entered geographical boundary information of cotton fields from the database, and simultaneously reads the shooting time and spatial coordinates of the multispectral images collected by the drone.

[0021] The center point coordinates of each extracted image are analyzed with the geographic boundaries of all small areas. Then, each image is assigned to a specific cotton field small area unit to establish the relationship between the image and the geographic area.

[0022] After determining the specific area to which each image belongs, the system uses its shooting timestamp as the query condition to automatically access external weather services and parse out the core fields representing weather phenomena. Based on preset rules, the system determines the weather condition as sunny or cloudy.

[0023] Perform data fusion to associate three dimensions of information for each image: specifically, associate the small area of ​​cotton field with the shooting time and the determined real-time weather conditions.

[0024] Furthermore, the panoramic spectral image generation module assesses the weather in each region, then utilizes the overlap of images captured by the drone to remove images with poor weather conditions, and selects the images with the best weather conditions to stitch together a panoramic view of the cotton field. When a cotton field area has no clear-sky images at any shooting time, the best version is directly selected from the available images, and the images are processed to make their visual and spectral characteristics resemble those of a clear-sky image. The process is as follows:

[0025] The system receives images that are associated with specific cotton field sub-regions and real-time weather conditions. It evaluates and marks the weather conditions of all images within each sub-region and establishes clear-sky images as the highest quality standard for priority selection.

[0026] Clear-sky imagery was chosen as the highest quality standard because its stable illumination and reliable spectral information are most conducive to accurate subsequent analysis.

[0027] By leveraging the overlap of images captured by drones, images covering the same area are compared, and images captured in cloudy weather are intelligently filtered and eliminated, thus initially selecting the image candidates with the best weather conditions for each small area.

[0028] Due to the characteristics of drone aerial photography, there will be some overlap between images taken at adjacent locations. This means that the same cotton field may be covered by multiple images taken at different times and under different weather conditions. Therefore, by utilizing this characteristic, it is possible to initially screen out the image candidates with the best weather conditions for each small area.

[0029] After completing the intelligent screening, an image with the best weather conditions is selected for each small area of ​​cotton field as the representative of the area. Then, these high-quality images selected from each area are seamlessly stitched together based on their common geographic coordinate information to generate a high-quality panoramic multispectral image covering the entire cotton field with consistent lighting conditions. The panoramic multispectral image is then processed by calling the processing methods for multispectral images under sunny conditions in historical data.

[0030] For certain cotton field areas, if no clear-sky images are captured during all flight shooting periods, an alternative plan is activated, which involves directly selecting the highest-resolution version from the available images for that area and processing the image.

[0031] Furthermore, by utilizing the overlap of images captured by drones to compare images covering the same area, and intelligently filtering and eliminating images captured under cloudy weather conditions, the preliminary selection of candidate images with the best weather conditions for each small area is as follows:

[0032] The system first calculates the actual geographical coverage of each image based on its precise GPS coordinates. Then, through spatial calculations, it automatically groups all images with overlapping areas into the same relation group, and all images in the group represent the same small cotton field area taken at different times.

[0033] After establishing a relation group for all overlapping images, the weather conditions of the images are extracted, and then the weather attributes are assigned to each image in the relation group, thereby clarifying the acquisition environment quality of each image in the group.

[0034] Each relationship group is analyzed independently, and the weather attributes of all images in the group are compared. Based on the rule of prioritizing clear sky images with better lighting conditions, all images marked as cloudy in the group are automatically filtered out and their status is marked as to be eliminated or as secondary selection, thus completing the initial quality screening.

[0035] After the comparison and filtering are completed, an optimized image candidate list is output for each geographic region, i.e., each relationship group. For relationship groups with clear sky images, the list will only contain high-quality clear sky images; for relationship groups with all images being cloudy, all images are retained as candidates and the region is marked as having no clear sky images.

[0036] Furthermore, the system first calculates the actual geographical coverage of each image based on its precise GPS coordinates, and then automatically groups all images with overlapping spatial areas into the same relation group through spatial calculations. The specific formula is as follows:

[0037] ;

[0038] in, This represents two different images that need to be determined to be overlapping. This represents the minimum east and minimum north coordinates among all points within the coverage rectangle of image A. This represents the largest east and largest north coordinates among all points within the coverage rectangle of image A. This represents the minimum east and minimum north coordinates among all points within the coverage rectangle of image B. This represents the largest east coordinate and the largest north coordinate among all points within the coverage rectangle of image B.

[0039] Furthermore, for certain cotton field areas, if no clear-sky images are captured during all flight shooting periods, an alternative plan is activated. This involves directly selecting the highest-resolution image from the available images for that area and processing it as follows:

[0040] When it is detected that all images of a specific small area of ​​cotton field are cloudy during all flight shooting periods, the alternative processing plan is automatically triggered.

[0041] For each image in the candidate set, a sharpness assessment is performed by calculating the gradient magnitude and edge intensity. Then, the assessment scores of all images are compared, and the image with the highest assessment score is selected as the image with the highest sharpness.

[0042] First, the image is processed using the methods for processing multispectral images under cloudy conditions from historical data. Then, an image enhancement algorithm is used to serialize this cloudy image so that its visual and spectral features are consistent with other clear-sky images.

[0043] Furthermore, the pest and disease identification and pesticide application prediction module, based on known panoramic multispectral imagery, uses a deep learning model to identify areas in the cotton field infected with pests and diseases, and quantitatively assesses the stress level of pests and diseases by analyzing the spectral reflectance characteristics of specific bands. The process for generating a variable pesticide application prescription map for the cotton field is as follows:

[0044] By inputting panoramic multispectral images into a pre-trained deep learning model, typical visual and spectral anomalies of pests and diseases are identified, and healthy areas and areas infected with pests and diseases in cotton fields are segmented to obtain spatial distribution information of pests and diseases.

[0045] For the identified infected areas, the reflectance characteristics of specific bands of vegetation are further analyzed, and the normalized vegetation index and plant senescence reflectance index are calculated. These are then compared with the standard values ​​of healthy areas to quantify the severity of pests and diseases. Multiple threshold intervals are then set, and the severity of pests and diseases in the infected areas is divided into three levels: mild, moderate, and severe.

[0046] The spatial distribution information and severity level information of pests and diseases are integrated, and a digital variable pesticide prescription map is generated for cotton fields according to the preset pesticide application rules.

[0047] This variable application prescription chart can directly guide spraying equipment on where and how much pesticide to apply, thereby achieving precise and efficient pest and disease control.

[0048] Compared with the prior art, the beneficial effects of the present invention are:

[0049] By setting up a data acquisition module to collect historical data on image processing methods under different weather conditions, key technical support can be provided for subsequent processes: In the panoramic spectral image generation module, historical clear-day image processing methods can be directly called to optimize the quality of the stitched panoramic image and ensure the reliability of its spectral information; when there are no clear-day images in some cotton field areas, historical cloudy image processing methods can be used in conjunction with image enhancement algorithms to adjust the selected cloudy images to clear-day features, ensuring the overall consistency of the panoramic image and also speeding up the efficiency of image processing;

[0050] By setting up an image alignment module, images can be aligned with shooting time, and cotton fields divided into small areas can be associated with real-time weather, thus effectively addressing the issue of weather changes during shooting. By analyzing image timestamps and spatial coordinates, and combining point and area analysis with cotton field divisions, the correspondence between images, small areas, and real-time weather can be accurately established, avoiding the situation where image and weather matching is chaotic due to weather fluctuations within the shooting time. This provides clear regional and weather basis for the subsequent panoramic spectral image generation module to select the best images, ensuring that the images selected for each small area meet the quality requirements.

[0051] By setting up a panoramic spectral image generation module, images with poor weather conditions can be removed through overlap analysis, and high-quality images can be selected and stitched together to form a panoramic image. On the one hand, the high-quality images selected by image overlap are themselves characterized by stable illumination and reliable spectral information, resulting in a more uniform quality panoramic image after stitching. On the other hand, panoramic images have advantages over processing individual images separately. The panoramic image as a whole utilizes the processing methods for multispectral images under clear weather conditions from historical data, eliminating the need to adjust parameters for each individual image. This significantly reduces processing steps and improves image processing efficiency.

[0052] When no clear-sky images are available for any of the shooting periods in the cotton field area, selecting the clearest version of the available images and processing it to reflect clear-sky characteristics can avoid the problem of missing data in that area due to weather limitations. By calculating the gradient amplitude and edge intensity, the clearest images are selected, and then combined with image enhancement algorithms to adjust their visual and spectral features. This ensures that the images of that area are consistent with other clear-sky images, ensuring the integrity and uniformity of the cotton field panorama. It also prevents the overall pest and disease identification results from being affected by poor image quality in some areas, thereby ensuring that the subsequently generated variable application prescription map can cover the entire cotton field, accurately match the pest and disease situation in each area, and achieve comprehensive and accurate pest and disease control in cotton fields. Attached Figure Description

[0053] Figure 1 This is a block diagram of a cotton field pest and disease control spraying dosage prediction system based on UAV multispectral remote sensing imagery, according to the present invention.

[0054] Figure 2 This is a flowchart of image screening in a cotton field pest and disease control spraying dosage prediction system based on UAV multispectral remote sensing imagery, according to the present invention. Detailed Implementation

[0055] The technical solution of the present invention will be clearly and completely described below with reference to the embodiments. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0056] like Figure 1 - Figure 2 As shown, a cotton field pest and disease control spraying amount prediction system based on UAV multispectral remote sensing imagery includes a data acquisition module, an image alignment module, a panoramic spectral image generation module, and a pest and disease identification and spraying amount prediction module.

[0057] The data acquisition module controls the drone to collect multispectral remote sensing images of cotton fields and uses GPS and IMU sensors to record the spatiotemporal location of each image. At the same time, it collects historical data on how the images were processed under different weather conditions.

[0058] Collecting historical data on image processing methods under different weather conditions can provide key technical support for the panoramic spectral image generation module. When stitching together panoramic cotton field images, the image processing methods under historical clear weather conditions can be directly called to optimize the quality of the panoramic image. When there are no clear weather images in some cotton field areas, the image processing methods under historical cloudy weather conditions can be used in combination with image enhancement algorithms to process the selected cloudy images so that their visual and spectral characteristics are consistent with those of clear weather images, ensuring that the overall quality of the generated panoramic multispectral image is uniform and the spectral information is reliable.

[0059] In this embodiment, the data acquisition module controls the drone to collect multispectral remote sensing images of the cotton field, and uses GPS and IMU sensors to record the spatiotemporal location of each image. Simultaneously, it collects historical data on image processing methods under different weather conditions. The processing procedure is as follows:

[0060] Based on the geographical boundaries of the cotton fields and preset flight parameters including flight altitude, flight speed and overlap rate, a complete data acquisition route is automatically planned for the UAV. The UAV is then controlled to fly automatically along the predetermined route, and its onboard multispectral camera is instructed to take pictures at designated spatial points, thereby systematically collecting multispectral remote sensing images covering the entire cotton field area.

[0061] At the moment the drone flies and the camera captures the image, the onboard GPS and IMU sensors are simultaneously triggered to record the latitude and longitude coordinates of the shooting center and the attitude angles of the drone platform at that time, including roll, pitch, and yaw.

[0062] Historical data is collected from the system database, and processing methods for multispectral images under different weather conditions, including sunny and cloudy conditions, are collected.

[0063] All multispectral images, along with their corresponding data and historical data processing methods, are integrated into a structured dataset.

[0064] The image alignment module aligns the images with the shooting time and divides the cotton field into multiple small areas based on the geographical information of the cotton field. At the same time, it associates the real-time weather conditions of each area to determine the location of the cotton field and the weather conditions corresponding to each image.

[0065] In this embodiment, the image alignment module aligns the images with the shooting time, divides the cotton field into multiple small areas based on the geographical information of the cotton field, and associates the real-time weather conditions of each area to determine the location of the cotton field and weather conditions corresponding to each image. The processing procedure is as follows:

[0066] The data of each multispectral remote sensing image file is analyzed, and the precise capture timestamp and spatial coordinates of each image are extracted to establish the correspondence between the image and the capture time.

[0067] The system retrieves pre-entered geographical boundary information of the cotton field and divides the entire cotton field into multiple small areas according to the set rules. Then, based on the image capture time, it queries and associates the real-time weather conditions of the area at that moment with external meteorological data. The weather conditions include sunny and cloudy. Each image is then accurately assigned to a specific small area of ​​the cotton field and marked with the corresponding capture time and detailed real-time weather conditions.

[0068] In this embodiment, pre-recorded geographical boundary information of the cotton field is invoked, and the entire cotton field is divided into multiple small areas according to set rules. Then, based on the image capture time, the real-time weather conditions matching the region at that moment are queried from external meteorological data. The weather conditions include sunny and cloudy. Each image is then accurately assigned to a specific small cotton field area and marked with the corresponding capture time. The detailed real-time weather condition processing process is as follows:

[0069] The system retrieves pre-entered geographical boundary information of cotton fields from the database, and simultaneously reads the shooting time and spatial coordinates of the multispectral images collected by the drone.

[0070] The center point coordinates of each extracted image are analyzed with the geographic boundaries of all small areas. Then, each image is assigned to a specific cotton field small area unit to establish the relationship between the image and the geographic area.

[0071] After determining the specific area to which each image belongs, the system uses its shooting timestamp as the query condition to automatically access external weather services and parse out the core fields representing weather phenomena. Based on preset rules, the system determines the weather condition as sunny or cloudy.

[0072] Perform data fusion operations to record three dimensions of information for each image, specifically linking the small area of ​​cotton field with the shooting time and the determined real-time weather conditions;

[0073] The panoramic spectral image generation module assesses the weather in each area, then uses drones to capture images with overlapping data to remove images with poor weather conditions, and selects the images with the best weather conditions to stitch together a panoramic view of the cotton field. When there are no clear-sky images of a cotton field area at any shooting time, the best version is directly selected from the available images and processed to make its visual and spectral characteristics resemble those of a clear-sky image.

[0074] In this embodiment, the panoramic spectral image generation module assesses the weather in each region, then uses the overlap of images captured by the drone to remove images with poor weather conditions, and selects the images with the best weather conditions to stitch together a panoramic view of the cotton field. When there are no clear-sky images of a certain cotton field area at any shooting time, the best version is directly selected from the available images, and the images are processed to make their visual and spectral characteristics resemble those of clear-sky images. The process is as follows:

[0075] The system receives images that are associated with specific cotton field sub-regions and real-time weather conditions. It evaluates and marks the weather conditions of all images within each sub-region and establishes clear-sky images as the highest quality standard for priority selection.

[0076] By leveraging the overlap of images captured by drones, images covering the same area are compared, and images captured in cloudy weather are intelligently filtered and eliminated, thus initially selecting the image candidates with the best weather conditions for each small area.

[0077] In this embodiment, the overlapping nature of images captured by drones is utilized to compare images covering the same area. Images captured under cloudy weather conditions are intelligently filtered and eliminated. The preliminary selection of candidate images with the best weather conditions for each small area is as follows:

[0078] The system first calculates the actual geographical coverage of each image based on its precise GPS coordinates. Then, through spatial calculations, it automatically groups all images with overlapping areas into the same relation group, and all images in the group represent the same small cotton field area taken at different times.

[0079] In this embodiment, the system first calculates the actual geographical coverage of each image based on its precise GPS coordinates, and then automatically groups all images with overlapping spatial areas into the same relation group through spatial calculations. The specific formula is as follows:

[0080] ;

[0081] in, This represents two different images that need to be determined to be overlapping. This represents the minimum east and minimum north coordinates among all points within the coverage rectangle of image A. This represents the largest east and largest north coordinates among all points within the coverage rectangle of image A. This represents the minimum east and minimum north coordinates among all points within the coverage rectangle of image B. This represents the largest east coordinate and the largest north coordinate among all points within the coverage rectangle of image B.

[0082] Since the overlap rate of drone images is usually set to 80% in the forward direction and 70% in the side direction, the actual geographical coverage of each image is calculated based on the precise GPS coordinates of each image. Then, through spatial calculation, images with overlapping areas in space are automatically grouped into the same relation group. The overlap rate of drone photos ensures that there are multiple images of the same cotton field area taken at different times, which can more efficiently select images under sunny conditions from the candidate images and finally stitch them together to form a complete image of the cotton field under sunny conditions.

[0083] The time required for drones to film an entire cotton field is usually determined by the specific area of ​​the field. When the time spent filming the cotton field is long, even if the weather conditions are known in advance, weather changes can still occur during the drone's flight, such as from sunny to cloudy. Directly stitching together images from different weather conditions may affect the accuracy of cotton field pest identification. At the same time, sunny images have the advantages of stable lighting and reliable spectral information. The stitched complete sunny image can avoid spectral interference from cloudy or other adverse weather images, ensuring uniform image quality across the entire cotton field. This provides accurate spatial distribution and spectral feature data for subsequent deep learning models for pest identification.

[0084] After establishing a relation group for all overlapping images, the weather conditions of the images are extracted, and then the weather attributes are assigned to each image in the relation group, thereby clarifying the acquisition environment quality of each image in the group.

[0085] Each relationship group is analyzed independently, and the weather attributes of all images in the group are compared. Based on the rule of prioritizing clear sky images with better lighting conditions, all images marked as cloudy in the group are automatically filtered out and their status is marked as to be eliminated or as secondary selection, thus completing the initial quality screening.

[0086] After the comparison and filtering are completed, an optimized image candidate list is output for each geographic region, i.e., each relationship group. For relationship groups with clear sky images, the list will only contain high-quality clear sky images; for relationship groups with all images being cloudy, all images are retained as candidates and the region is marked as having no clear sky images.

[0087] After completing the intelligent screening, an image with the best weather conditions is selected for each small area of ​​cotton field as the representative of the area. Then, these high-quality images selected from each area are seamlessly stitched together based on their common geographic coordinate information to generate a high-quality panoramic multispectral image covering the entire cotton field with consistent lighting conditions. The panoramic multispectral image is then processed by calling the processing methods for multispectral images under sunny conditions in historical data.

[0088] For certain cotton field areas, if no clear-sky images are captured during all flight shooting periods, an alternative plan is activated: the highest-resolution image is directly selected from the available images for that area, and then processed.

[0089] The pest and disease identification and spraying amount prediction module uses a deep learning model to identify areas in cotton fields infected with pests and diseases based on known panoramic multispectral images. By analyzing the spectral reflectance characteristics of specific bands, it quantitatively assesses the stress level of pests and diseases and generates variable application prescription maps for cotton fields.

[0090] In this embodiment, the pest and disease identification and pesticide application prediction module, based on known panoramic multispectral imagery, uses a deep learning model to identify areas in the cotton field infected with pests and diseases. It also analyzes the spectral reflectance characteristics of specific bands to quantitatively assess the stress level of pests and diseases, generating a variable pesticide application prescription map for the cotton field. The processing steps are as follows:

[0091] By inputting panoramic multispectral images into a pre-trained deep learning model, typical visual and spectral anomalies of pests and diseases are identified, and healthy areas and areas infected with pests and diseases in cotton fields are segmented to obtain spatial distribution information of pests and diseases.

[0092] For the identified infected areas, the reflectance characteristics of specific bands of vegetation are further analyzed, and the normalized vegetation index and plant senescence reflectance index are calculated. These are then compared with the standard values ​​of healthy areas to quantify the severity of pests and diseases. Multiple threshold intervals are then set, and the severity of pests and diseases in the infected areas is divided into three levels: mild, moderate, and severe.

[0093] The spatial distribution information and severity level information of pests and diseases are integrated, and a digital variable pesticide prescription map is generated for cotton fields according to the preset pesticide application rules.

[0094] The above embodiments are only used to illustrate the technical methods of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical methods of the present invention without departing from the spirit and scope of the technical methods of the present invention.

Claims

1. A system for predicting pesticide application rates for cotton field pest and disease control based on UAV multispectral remote sensing imagery, characterized in that, It includes a data acquisition module, an image alignment module, a panoramic spectral image generation module, and a pest and disease identification and spraying dosage prediction module; The data acquisition module controls the drone to collect multispectral remote sensing images of cotton fields, and uses GPS and IMU sensors to record the spatiotemporal location of each image. At the same time, it collects historical data on the processing methods of images under different weather conditions. The image alignment module aligns the images with the shooting time, divides the cotton field into multiple small areas based on the geographical information of the cotton field, and associates the real-time weather conditions of each area to determine the location of the cotton field and weather conditions corresponding to each image. The panoramic spectral image generation module assesses the weather in each region, then uses the overlap of images taken by drones to remove images with poor weather conditions, and selects the images with the best weather conditions to stitch together a panoramic view of the cotton field. When there are no clear weather images for a certain cotton field area at any shooting time, the best version is directly selected from the available images and processed to make its visual and spectral characteristics resemble those of a clear weather image. The pest and disease identification and pesticide application prediction module uses a deep learning model to identify areas in cotton fields infected with pests and diseases based on known panoramic multispectral images. By analyzing the spectral reflectance characteristics of specific bands, it quantifies and assesses the stress level of pests and diseases, generating variable pesticide application prescription maps for cotton fields.

2. The cotton field pest and disease control spraying dosage prediction system based on UAV multispectral remote sensing imagery according to claim 1, characterized in that, The data acquisition module controls the drone to collect multispectral remote sensing images of the cotton field, and uses GPS and IMU sensors to record the spatiotemporal location of each image. It also collects historical data on image processing methods under different weather conditions. The processing steps are as follows: Based on the geographical boundaries of the cotton fields and preset flight parameters including flight altitude, flight speed and overlap rate, a complete data acquisition route is automatically planned for the UAV. The UAV is then controlled to fly automatically along the predetermined route, and its onboard multispectral camera is instructed to take pictures at designated spatial points, thereby systematically collecting multispectral remote sensing images covering the entire cotton field area. At the moment the drone flies and the camera captures the image, the onboard GPS and IMU sensors are simultaneously triggered to record the latitude and longitude coordinates of the shooting center and the attitude angles of the drone platform at that time, including roll, pitch, and yaw. Historical data is collected from the system database, and processing methods for multispectral images under different weather conditions, including sunny and cloudy conditions, are collected. All multispectral images, along with their corresponding data and historical data processing methods, are integrated into a structured dataset.

3. The cotton field pest and disease control spraying dosage prediction system based on UAV multispectral remote sensing imagery according to claim 1, characterized in that, The image alignment module aligns the images with the shooting time, divides the cotton field into multiple small areas based on the geographical information of the cotton field, and associates the real-time weather conditions of each area to determine the location of the cotton field and weather conditions corresponding to each image. The processing procedure is as follows: The data of each multispectral remote sensing image file is analyzed, and the precise capture timestamp and spatial coordinates of each image are extracted to establish the correspondence between the image and the capture time. The system retrieves pre-entered geographical boundary information of the cotton field and divides the entire cotton field into multiple small areas according to the set rules. Then, based on the image capture time, it queries and associates the real-time weather conditions of the area at that moment with external meteorological data. The weather conditions include sunny and cloudy. Each image is then accurately assigned to a specific small area of ​​the cotton field and marked with the corresponding capture time and detailed real-time weather conditions.

4. The cotton field pest and disease control spraying dosage prediction system based on UAV multispectral remote sensing imagery according to claim 3, characterized in that, The process involves calling pre-recorded geographical boundary information of the cotton field and dividing the entire cotton field into multiple small areas according to set rules. Then, based on the image capture time, it queries and associates the real-time weather conditions of the area at that moment with external meteorological data. The weather conditions include sunny and cloudy. Each image is then precisely assigned to a specific small cotton field area and marked with its corresponding capture time. The detailed real-time weather processing process is as follows: The system retrieves pre-entered geographical boundary information of cotton fields from the database, and simultaneously reads the shooting time and spatial coordinates of the multispectral images collected by the drone. The center point coordinates of each extracted image are analyzed with the geographic boundaries of all small areas. Then, each image is assigned to a specific cotton field small area unit to establish the relationship between the image and the geographic area. After determining the specific area to which each image belongs, the system uses its shooting timestamp as the query condition to automatically access external weather services and parse out the core fields representing weather phenomena. Based on preset rules, the system determines the weather condition as sunny or cloudy. Perform data fusion to associate three dimensions of information for each image: specifically, associate the small area of ​​cotton field with the shooting time and the determined real-time weather conditions.

5. The cotton field pest and disease control spraying dosage prediction system based on UAV multispectral remote sensing imagery according to claim 1, characterized in that, The panoramic spectral image generation module assesses the weather in each region, then uses drone-captured images to eliminate those with poor weather conditions, and selects the images with the best weather to stitch together a panoramic view of the cotton field. When a cotton field area has no clear-sky images at any shooting time, the best version is directly selected from the available images, and the images are processed to make their visual and spectral characteristics resemble those of a clear-sky image. The process is as follows: The system receives images that are associated with specific cotton field sub-regions and real-time weather conditions. It evaluates and marks the weather conditions of all images within each sub-region and establishes clear-sky images as the highest quality standard for priority selection. By leveraging the overlap of images captured by drones, images covering the same area are compared, and images captured in cloudy weather are intelligently filtered and eliminated, thus initially selecting the image candidates with the best weather conditions for each small area. After completing the intelligent screening, an image with the best weather conditions is selected for each small area of ​​cotton field as the representative of the area. Then, these high-quality images selected from each area are seamlessly stitched together based on their common geographic coordinate information to generate a high-quality panoramic multispectral image covering the entire cotton field with consistent lighting conditions. The panoramic multispectral image is then processed by calling the processing methods for multispectral images under sunny conditions in historical data. For certain cotton field areas, if no clear-sky images are captured during all flight shooting periods, an alternative plan is activated, which involves directly selecting the highest-resolution version from the available images for that area and processing the image.

6. The cotton field pest and disease control spraying dosage prediction system based on UAV multispectral remote sensing imagery according to claim 4, characterized in that, The process of utilizing the overlap of images captured by drones to compare images covering the same area, intelligently filtering and eliminating images taken under cloudy weather conditions, and initially selecting candidate images with the best weather conditions for each small area is as follows: The system first calculates the actual geographical coverage of each image based on its precise GPS coordinates. Then, through spatial calculations, it automatically groups all images with overlapping areas into the same relation group, and all images in the group represent the same small cotton field area taken at different times. After establishing a relation group for all overlapping images, the weather conditions of the images are extracted, and then the weather attributes are assigned to each image in the relation group, thereby clarifying the acquisition environment quality of each image in the group. Each relationship group is analyzed independently, and the weather attributes of all images in the group are compared. Based on the rule of prioritizing clear sky images with better lighting conditions, all images marked as cloudy in the group are automatically filtered out and their status is marked as to be eliminated or as secondary selection, thus completing the initial quality screening. After the comparison and filtering are completed, an optimized image candidate list is output for each geographic region, i.e., each relationship group. For relationship groups with clear sky images, the list will only contain high-quality clear sky images; for relationship groups with all images being cloudy, all images are retained as candidates and the region is marked as having no clear sky images.

7. A cotton field pest and disease control spraying dosage prediction system based on UAV multispectral remote sensing imagery according to claim 6, characterized in that, The system first calculates the actual geographical coverage of each image based on its precise GPS coordinates, and then automatically groups all images with overlapping spatial areas into the same relation group through spatial calculations. The specific formula is as follows: ； in, This represents two different images that need to be determined to be overlapping. This represents the minimum east and minimum north coordinates among all points within the coverage rectangle of image A. This represents the largest east and largest north coordinates among all points within the coverage rectangle of image A. This represents the minimum east and minimum north coordinates among all points within the coverage rectangle of image B. This represents the largest east coordinate and the largest north coordinate among all points within the coverage rectangle of image B.

8. A cotton field pest and disease control spraying dosage prediction system based on UAV multispectral remote sensing imagery according to claim 5, characterized in that, For certain cotton field areas, if no clear-sky images are captured during all flight shooting periods, an alternative plan is activated. This involves directly selecting the highest-resolution image from the available images for that area and processing it as follows: When it is detected that all images of a specific small area of ​​cotton field are cloudy during all flight shooting periods, the alternative processing plan is automatically triggered. For each image in the candidate set, a sharpness assessment is performed by calculating the gradient magnitude and edge intensity. Then, the assessment scores of all images are compared, and the image with the highest assessment score is selected as the image with the highest sharpness. First, the image is processed using the methods for processing multispectral images under cloudy conditions from historical data. Then, an image enhancement algorithm is used to serialize this cloudy image so that its visual and spectral features are consistent with other clear-sky images.

9. A cotton field pest and disease control spraying dosage prediction system based on UAV multispectral remote sensing imagery according to claim 1, characterized in that, The pest and disease identification and pesticide application prediction module, based on known panoramic multispectral imagery, uses a deep learning model to identify areas in cotton fields infected with pests and diseases. It also analyzes the spectral reflectance characteristics of specific bands to quantitatively assess the stress level of pests and diseases, generating variable pesticide application prescription maps for the cotton fields. The processing steps are as follows: By inputting panoramic multispectral images into a pre-trained deep learning model, typical visual and spectral anomalies of pests and diseases are identified, and healthy areas and areas infected with pests and diseases in cotton fields are segmented to obtain spatial distribution information of pests and diseases. For the identified infected areas, the reflectance characteristics of specific bands of vegetation are further analyzed, and the normalized vegetation index and plant senescence reflectance index are calculated. These are then compared with the standard values ​​of healthy areas to quantify the severity of pests and diseases. Multiple threshold intervals are then set, and the severity of pests and diseases in the infected areas is divided into three levels: mild, moderate, and severe. The spatial distribution information and severity level information of pests and diseases are integrated, and a digital variable pesticide prescription map is generated for cotton fields according to the preset pesticide application rules.

Images