Camellia oleifera optimal disease strain automatic screening method and system based on unmanned aerial vehicle image
By using multispectral images from drones and deep learning models to identify the canopy of Camellia oleifera trees, combined with high-precision positioning technology, the problems of low efficiency and poor accuracy in screening superior and diseased Camellia oleifera trees have been solved, achieving efficient and accurate screening and management.
Patent Information
- Application Number
- CN202610831164.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Filing Date
- 2026-06-10
- Publication Date
- 2026-07-10
AI Technical Summary
In existing technologies, the screening of superior and diseased Camellia oleifera plants relies on manual labor, which is inefficient and inaccurate, and cannot accurately locate the plants. Furthermore, existing drone monitoring technologies have limited data sources and lack geographic information, failing to form a closed-loop solution.
Aerial photography is conducted using drones equipped with multispectral cameras and high-precision positioning modules to acquire multispectral and visible light images. Tree canopies are identified through deep learning models, vegetation indices are calculated for filtering, and accurate geographical locations are output.
The entire process of screening superior and diseased Camellia oleifera plants has been automated, improving screening efficiency and accuracy. The positioning accuracy has reached the sub-meter level, providing a precise basis for management decisions.
Smart Images
Figure CN122368798A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of forestry remote sensing and precision forestry technology, specifically to an automatic screening method and system for superior and diseased Camellia oleifera plants based on UAV imagery. Background Technology
[0002] Camellia oleifera is an important woody oilseed tree species in my country, and the healthy development of its industry is of great significance for ensuring food and oil security and promoting economic development in mountainous areas. In the refined management and breeding of superior varieties of Camellia oleifera forests, quickly and accurately screening out the "superior trees" with excellent growth conditions and the "disease trees" suffering from pests, diseases, and adverse stresses from large-scale plantations is the basic prerequisite for implementing precision forestry measures (such as targeted fertilization and targeted spraying) and carrying out genetic improvement.
[0003] For a long time, this screening work has mainly relied on manual field surveys. Technicians need to go deep into the forest area and visually observe the phenotypic characteristics of each camellia tree, such as canopy color, foliage density, and fruit condition, judging its health status based on personal experience, and manually recording (or roughly marking) the locations of suspected diseased or superior trees using handheld GPS devices. This traditional method has revealed several inherent and insurmountable shortcomings in long-term practice: Low efficiency and scalability: Manual patrols are time-consuming and labor-intensive. It often takes several days to survey a camellia oleifera forest covering hundreds of acres, which cannot meet the timeliness requirements of modern large-scale and intensive forestry management, and is even more difficult to apply to the survey of large-scale planting bases covering tens of thousands of acres.
[0004] Highly subjective, with poor accuracy and consistency: Judgments are heavily reliant on the investigator's personal experience and subjective standards. Differences can exist between different technicians, and even among the same technician at different times, leading to low repeatability. Statistics show that relying solely on visual judgment typically yields an accuracy rate of only around 70%, with high rates of misjudgment and missed judgment.
[0005] Lack of objective quantitative indicators: Manual judgment remains at the qualitative or semi-quantitative level, failing to provide continuous and accurate quantitative assessment of plant health status. This makes it difficult to conduct scientific statistical analysis of the screening results and to provide stable data support for subsequent management decisions.
[0006] Insufficient positioning accuracy hinders precise operations: Manually recorded location information typically has large errors (often exceeding 5 meters) or only marks small areas. This level of accuracy cannot support subsequent precise operations; for example, it cannot guide drones or automated equipment to accurately apply pesticides to individual diseased plants, reducing the effectiveness of management measures and potentially increasing costs.
[0007] In recent years, unmanned aerial vehicle (UAV) remote sensing technology has been widely used in forestry monitoring due to its flexibility and efficiency. Some studies have attempted to use UAVs equipped with visible light cameras for crop growth monitoring. However, when applied to the specific scenario of automated screening of superior and diseased camellia oleifera plants, existing technologies still have significant limitations: The data source is singular, and the sensitivity of health assessment is insufficient: it relies solely on visible light (RGB) imagery, primarily utilizing color and texture features. However, in the early stages of disease or stress in Camellia oleifera, the canopy color may not show significant changes in the visible light band, but internal physiological parameters such as photosynthetic efficiency and water status have already altered. This information is mainly reflected in multispectral bands such as near-infrared and red-edge. Using only visible light data makes it difficult to accurately identify physiological stress in the early stages, resulting in a low upper limit for screening accuracy (typically below 80%).
[0008] Missing or inaccurate geographic information: Many solutions do not integrate high-precision positioning modules (such as RTK) or only use consumer-grade GNSS, resulting in inaccurate or completely missing geographic location information in the generated images. This makes it impossible to correlate the "diseases" analyzed from the images with their precise real-world locations, making it difficult to apply the automated screening results. In essence, only "identification" is achieved, but "location" is not.
[0009] The technical processes are fragmented and a closed-loop solution has not been formed: existing methods either focus on canopy segmentation or vegetation index calculation, and fail to organically integrate a series of key technical links such as high-precision geographic information acquisition, high-precision single tree canopy extraction, multispectral-based physiological health quantitative assessment, and spatial coordinate output of results into a complete automated workflow.
[0010] Therefore, in current forestry production practices, there is an urgent need for a method to screen superior and diseased Camellia oleifera plants that can integrate multi-source remote sensing data, integrate high-precision positioning technology, and realize full automation from data acquisition to precise coordinate output. This method would overcome the drawbacks of manual screening and existing rudimentary drone monitoring technologies, and achieve efficient, objective, and accurate screening and precise management of forest tree phenotypes. Summary of the Invention
[0011] The technical problem to be solved by this invention is to address the issues of low efficiency, poor accuracy, and inability to accurately locate superior and diseased Camellia oleifera plants in the existing technology, which rely on manual screening. This invention provides an automatic screening method and system for superior and diseased Camellia oleifera plants based on UAV imagery.
[0012] To solve the above-mentioned technical problems, the technical solution adopted by the method of the present invention includes the following steps: S1, Data Acquisition and Stitching: Control a drone equipped with a multispectral camera, a visible light camera and a high-precision positioning module to take aerial photos of the target camellia oleifera forest, acquire multispectral images and visible light images with geographic coordinates, and stitch the multispectral images and visible light images to generate an orthophoto map with geographic reference information. S2, Canopy Extraction: Based on the orthophoto image obtained in step S1, multiple camellia tree canopies in the image are identified and located through a pre-trained camellia tree canopy target detection model, and the pixel position information of each canopy is obtained. The camellia tree canopy target detection model is a model based on a deep learning framework. S3, Health Assessment and Screening: Based on the multispectral image data acquired in S1, calculate the growth assessment index reflecting the physiological state of each canopy identified in S2; statistically sort the growth assessment index values of all canopies, and automatically screen out the canopy subsets characterized as superior and diseased plants according to preset rules. S4. Output Results: Using the georeferenced information included in the TIFF file, batch conversion from pixel coordinates to latitude and longitude is achieved, and the filtered results containing the geographical location information of the canopies of the superior and diseased trees are output. These results are then imported into the GIS system for precise management.
[0013] As a further improvement to the method of this invention, in step S1, the multispectral camera covers at least two non-visible light bands, including the near-infrared band; the orthophoto image is a GeoTIFF format file containing geographic coordinates and multiple spectral band information.
[0014] As a further improvement to the method of this invention, step S2 specifically includes: S21, Image slicing: The orthophoto image or its derived visible light image is segmented to generate multiple slice images; S22, Model Recognition: Input the sliced image into a pre-trained Camellia oleifera tree canopy target detection model, and automatically output the pixel coordinates of the outer frame of each canopy; S23, Result Correction: Perform manual or automated post-processing on the results of automatic model recognition to delete or correct incorrectly identified non-canopy targets.
[0015] As a further improvement to the method of this invention, the target detection model for the camellia tree canopy is the YOLO model.
[0016] As a further improvement to the method of this invention, in step S3, the growth assessment index is a vegetation index, which is calculated based on the reflectance of the near-infrared band and the red band in the multispectral image.
[0017] As a further improvement to the method of this invention, in step S3, the multispectral image data is used to calculate the average vegetation index of the canopy of each camellia tree. The indices are sorted from smallest to largest, and the 3% of plants with the lowest indices are taken as diseased plants, while the 3% of plants with the highest indices are taken as superior plants. The outer frame pixel coordinates of the diseased plants and the superior plants are output respectively. The average vegetation index includes GVI and NDVI, and its calculation formula is as follows: GVI = (NIR + Red) / (NIR - Red) NDVI = (NIR - Red) / (NIR + Red) NIR stands for near-infrared band, and Red stands for red light band.
[0018] As a further improvement to the method of this invention, the growth assessment index may also be any one of the following: normalized difference vegetation index, greenness vegetation index, enhanced vegetation index, or soil-regulated vegetation index.
[0019] As a further improvement to the method of this invention, in step S3, the preset rule is: the tree canopy that is in the highest value range in the growth assessment index ranking and accounts for a first predetermined proportion is judged as a superior tree; the tree canopy that is in the lowest value range and accounts for a second predetermined proportion is judged as a diseased tree, wherein the first predetermined proportion and the second predetermined proportion are equal or unequal.
[0020] As a further improvement to the method of this invention, step S4 includes: S41, Coordinate Transformation: Based on the geographic reference information of the orthophoto map, the pixel coordinates of the center points of the crowns of the superior and diseased trees selected in step S3 are converted into latitude and longitude coordinates. S42, Formatted Output: Output the converted coordinate information, along with the tree crown identifier and growth assessment index value, as a structured data file or database record. The formula for calculating the pixel coordinates of the center point of the tree canopy outline is: X coordinate = (top left X + bottom right X) / 2, Y coordinate = (top left Y + bottom right Y) / 2.
[0021] This invention also provides an automatic screening system for superior and diseased Camellia oleifera plants based on UAV imagery, to implement the above-mentioned automatic screening method for superior and diseased Camellia oleifera plants based on UAV imagery, comprising: The data acquisition unit includes an unmanned aerial vehicle (UAV) platform, as well as a multispectral imaging device, a visible light imaging device, and a high-precision positioning device mounted on it, for acquiring image data of camellia oleifera forests with geographic coordinates. A data processing and analysis unit, communicatively connected to the data acquisition unit, includes at least one processor and a memory. The memory stores a computer program, which, when executed by the processor, performs the functions of the following modules: The image processing module is used to stitch together the acquired image data to generate orthophoto maps; The canopy recognition module is used to automatically identify and locate the canopy of camellia oleifera trees from images; The health assessment module is used to calculate the growth assessment index of the tree canopy based on multispectral data, and automatically screen the superior trees and diseased trees according to the index ranking. The results generation module is used to convert the pixel coordinates of the filtered results into geographic coordinates and output them.
[0022] Compared with the prior art, the advantages of the present invention are as follows: This invention automates the entire process from data acquisition to result output, greatly improving screening efficiency and saving significant manpower. By using vegetation indices calculated from multispectral data as quantitative health indicators, subjective judgment is avoided, resulting in high screening accuracy. Combined with RTK centimeter-level positioning, the geographic location accuracy of the screening results reaches sub-meter level, far exceeding manual recording. Furthermore, the output result file contains precise latitude and longitude coordinates, which can be directly imported into a forestry GIS system, providing direct and scientific decision-making basis for subsequent precise forestry operations (such as variable-rate pesticide application and individual tree management). Attached Figure Description
[0023] Figure 1 This is a basic flowchart of the present invention. Detailed Implementation
[0024] The present invention will be further described below with reference to the accompanying drawings and specific preferred embodiments, but this does not limit the scope of protection of the present invention.
[0025] The technical solution adopted in this embodiment includes the following steps: like Figure 1 As shown, the overall solution in this example is to use a drone equipped with a multispectral camera and a visible light camera to take aerial orthophotos of the camellia oleifera forest. Based on the visible light orthophotos, the accurate pixel coordinates of the camellia oleifera tree crowns in the image data are extracted. Then, based on the extracted pixel coordinates of the camellia oleifera tree crowns, the growth status of the corresponding pixel range of the multispectral image is analyzed. The growth status analysis results are statistically sorted, and the geographical information of superior and diseased trees is output. Finally, the results are output.
[0026] S1, Data Acquisition and Stitching: Control a drone equipped with a multispectral camera, a visible light camera and a high-precision positioning module to take aerial photos of the target camellia oleifera forest, acquire multispectral images and visible light images with geographic coordinates, and stitch the multispectral images and visible light images to generate an orthophoto map with geographic reference information. S2, Canopy Extraction: Based on the orthophoto image obtained in step S1, multiple camellia tree canopies in the image are identified and located through a pre-trained camellia tree canopy target detection model, and the pixel position information of each canopy is obtained. The camellia tree canopy target detection model is a model based on a deep learning framework. S3, Health Assessment and Screening: Based on the multispectral image data acquired in S1, calculate the growth assessment index reflecting the physiological state of each canopy identified in S2; statistically sort the growth assessment index values of all canopies, and automatically screen out the canopy subsets characterized as superior and diseased plants according to preset rules. S4. Output Results: Using the georeferenced information included in the TIFF file, batch conversion from pixel coordinates to latitude and longitude is achieved, and the filtered results containing the geographical location information of the canopies of the superior and diseased trees are output. These results are then imported into the GIS system for precise management.
[0027] In a specific application example, in step S1, the multispectral camera covers at least two non-visible light bands, including the near-infrared band; the orthophoto is a GeoTIFF format file containing geographic coordinates and multiple spectral band information.
[0028] In a specific application example, step S2 specifically includes: S21, Image slicing: The orthophoto image or its derived visible light image is segmented to generate multiple slice images; S22, Model Recognition: Input the sliced image into a pre-trained Camellia oleifera tree canopy target detection model, and automatically output the pixel coordinates of the outer frame of each canopy; S23, Result Correction: Perform manual or automated post-processing on the results of automatic model recognition to delete or correct incorrectly identified non-canopy targets.
[0029] In a specific application example, the target detection model for the camellia tree canopy is the YOLO model.
[0030] In a specific application example, step S3 involves using multispectral image data to calculate the average vegetation index of each camellia tree canopy. The indices are sorted from smallest to largest, with the bottom 3% of trees identified as diseased and the top 3% as superior. The outer frame pixel coordinates of the diseased and superior trees are output separately. The average vegetation index includes GVI and NDVI, and its calculation formula is as follows: GVI = (NIR + Red) / (NIR - Red) NDVI = (NIR - Red) / (NIR + Red) NIR stands for near-infrared band, and Red stands for red light band.
[0031] In a specific application example, in step S3, the growth assessment index is a vegetation index, which is calculated based on the reflectance of the near-infrared band and the red band in the multispectral image.
[0032] In specific application examples, the vegetation index is any one of the following: Normalized Difference Vegetation Index (NDVI), Greenness Vegetation Index (GVI), Enhanced Vegetation Index (EVI), or Soil-Regulated Vegetation Index (SAVI).
[0033] In a specific application example, in step S3, the preset rule is: the tree canopy that is in the highest value range in the growth assessment index ranking and accounts for a first predetermined proportion is judged as a superior tree; the tree canopy that is in the lowest value range and accounts for a second predetermined proportion is judged as a diseased tree, wherein the first predetermined proportion and the second predetermined proportion are equal or unequal.
[0034] In a specific application example, step S4 includes: S41, Coordinate Transformation: Based on the geographic reference information of the orthophoto map, the pixel coordinates of the center points of the crowns of the superior and diseased trees selected in step S3 are converted into latitude and longitude coordinates. S42, Formatted Output: Output the converted coordinate information, along with the tree crown identifier and growth assessment index value, as a structured data file or database record. The formula for calculating the pixel coordinates of the center point of the tree canopy outline is: X coordinate = (top left X + bottom right X) / 2, Y coordinate = (top left Y + bottom right Y) / 2.
[0035] In this embodiment, the specific implementation steps are as follows: Step S1, Data Acquisition and Assembly: This step aims to acquire multi-source remote sensing imagery data with high-precision geographic information. A DJI Matrice M300 RTK drone was selected as the flight platform. This drone was equipped with a RedEdge-MX multispectral camera (its spectral bands include blue, green, red, red-edge, and near-infrared bands) and a Zenmuse P1 full-frame visible light camera. Before the flight mission, an automatic flight path was planned using ground station software to ensure full coverage aerial photography of the target camellia oleifera forest (approximately 150 acres). Key flight parameter settings were as follows: relative altitude set at 70 meters, forward overlap at 80%, and lateral overlap at 75%, to ensure the quality of subsequent image modeling. Throughout the aerial photography process, the drone's RTK module continuously received differential signals, injecting precise latitude, longitude, elevation, and attitude information into each captured image, achieving centimeter-level (typically 2-3 centimeters) positioning accuracy.
[0036] After aerial photography, the acquired raw images with RTK location information were imported into the professional image processing software Pix4Dmapper. The software first performed aerial triangulation on the multispectral and visible light images separately, then fused them using high-precision positioning information to generate an orthophoto map with high spatial resolution, true color information, and multispectral information, along with accurate georeferenced data. The output was in GeoTIFF format, with a geographic coordinate system of WGS84 UTM 50N and a ground sampling distance (GSD) of 2.5 cm / pixel. This file serves as the foundation for all subsequent processing using this method.
[0037] Step S2, Canopy Extraction: The goal of this step is to efficiently and accurately identify the canopy of each individual camellia tree from a large-scale orthophoto. First, the high-resolution visible light composite image generated in step S1 (or the RGB bands extracted directly from GeoTIFF) is sliced. Using a Python script in conjunction with the GDAL library, the entire image is uniformly cropped into small 1024x1024 pixel patches. This operation aims to adapt to the input size of subsequent deep learning models and improve batch processing efficiency.
[0038] Next, canopy target recognition is performed. The aforementioned slice images are input into a pre-trained deep learning model for camellia tree canopy detection. This model uses the YOLOv8s architecture and is trained using a large number of labeled visible light images of camellia forests, enabling it to effectively identify camellia tree canopies in complex backgrounds (such as those with weeds, bare soil, and shadows). The model infers for each slice, outputting the bounding boxes of all detected camellia tree canopies, and records the relative pixel coordinates (top left corner X1, Y1, bottom right corner X2, Y2) and confidence score of each bounding box within the slice image in JSON format.
[0039] Finally, result correction and coordinate integration are performed. The identification results of all slices are summarized, and based on the slice's position in the original image, the coordinates of the canopy bounding boxes are uniformly transformed to the global pixel coordinate system of the original orthophoto image. Although the model has a high accuracy rate (over 92%), there may still be cases where dense weeds or shadowy patches are misidentified as tree canopies. Therefore, an operator is assigned to quickly review the automatic identification results using a visualization tool and manually delete obviously erroneous detection boxes. This process can be completed within 15-30 minutes, ultimately ensuring that the overall canopy extraction accuracy is no less than 99%, and obtaining a list containing the global pixel coordinates of all camellia tree canopies in the forest area.
[0040] Step S3, Health Assessment and Screening: This step utilizes multispectral data to quantitatively assess the physiological state of each identified camellia oleifera tree canopy and automatically classify it accordingly. First, the vegetation index is calculated. Based on the multispectral GeoTIFF file generated in step S1, reflectance data in the near-infrared (NIR) and red bands are read. For each canopy bounding box obtained in step S2, the average reflectance value of all pixels within the box in these two bands is calculated. Subsequently, the normalized differential vegetation index is calculated for each tree using the following formula: NDVI = (NIR - Red) / (NIR + Red); The NDVI value ranges from -1 to 1. For healthy green vegetation, it is usually greater than 0. The higher the value, the more lush the vegetation, the greater the biomass, and the stronger the photosynthesis.
[0041] Next, statistical sorting and automatic filtering are performed. The NDVI values of all camellia oleifera tree canopies within the forest area are sorted from highest to lowest. In this embodiment, the filtering threshold ratio is set to the top 3% and bottom 3%. That is, the plants with the highest NDVI values in the top 3% are automatically marked as "superior plants," representing the best-growing and healthiest individuals in the area; the plants with the lowest NDVI values in the bottom 3% are automatically marked as "disease plants," representing individuals that may be suffering from stresses such as pests, diseases, drought, and malnutrition. The global pixel coordinates of the canopy bounding boxes corresponding to these superior and disease plants are recorded. It can be understood that the 3% ratio can be adjusted within the range of 1% to 5% according to actual management needs.
[0042] Step S4, Result Transformation and Output: This step transforms the image filtering results into usable geographic information in the field. First, coordinate transformation is performed. For each tree canopy marked as superior or diseased, the center pixel coordinates (X_center, Y_center) of its bounding box are calculated. Then, using the geographic transformation parameters provided in the GeoTIFF file from step S1, the center pixel coordinates are batch-converted to their corresponding latitude and longitude coordinates (longitude, latitude) through affine transformation.
[0043] Finally, format the output results. Generate a structured data file from the final screening results. For example, create a tabular file named "Oil_Camellia_Screening_Result.csv" with columns including: serial number, plant type (superior / disease-prone), longitude, latitude, and NDVI value. This file format is universal and can be directly imported into geographic information system software such as ArcGIS and QGIS, or precision forestry management platforms. Users can open this file in GIS software, and the selected superior and disease-prone plants will be accurately displayed as point layers on a graphical location coordinate interface, which can then be directly used to guide subsequent forestry management operations such as field inspections, precision fertilization, and targeted pesticide application.
[0044] This specific implementation demonstrates how to solve the data source and positioning problems through "UAV multispectral / visible light fusion acquisition + RTK positioning," how to solve the single-tree segmentation problem through "visible light-based deep learning-based canopy extraction," how to solve the objective quantitative classification problem through "health assessment and statistical screening based on multispectral NDVI calculation," and finally how to solve the problem of application of results through "pixel-geographic coordinate transformation." This method transforms the traditional manual, subjective, and inefficient screening process into an automated, objective, efficient, and high-precision technical workflow, increasing the screening accuracy to over 90% and achieving sub-meter positioning accuracy. It provides a complete technical solution for the precise management of large areas of camellia oleifera and other potentially economic forests.
[0045] This embodiment also includes an automatic screening system for superior and diseased Camellia oleifera plants based on UAV imagery, used to implement the above-mentioned automatic screening method for superior and diseased Camellia oleifera plants based on UAV imagery, including: The data acquisition unit includes an unmanned aerial vehicle (UAV) platform, as well as a multispectral imaging device, a visible light imaging device, and a high-precision positioning device mounted on it, for acquiring image data of camellia oleifera forests with geographic coordinates. A data processing and analysis unit, communicatively connected to the data acquisition unit, includes at least one processor and a memory. The memory stores a computer program, which, when executed by the processor, performs the functions of the following modules: The image processing module is used to stitch together the acquired image data to generate orthophoto maps; The canopy recognition module is used to automatically identify and locate the canopy of camellia oleifera trees from images; The health assessment module is used to calculate the growth assessment index of the tree canopy based on multispectral data, and automatically screen the superior trees and diseased trees according to the index ranking. The results generation module is used to convert the pixel coordinates of the filtered results into geographic coordinates and output them.
[0046] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-readable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code. This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create a machine for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to operate in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The functions specified in one or more boxes. These computer program instructions may also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable apparatus for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.
[0047] The above description is merely a preferred embodiment of the present invention. The scope of protection of the present invention is not limited to the above embodiments. All technical solutions falling within the scope of the present invention's concept are within the scope of protection of the present invention. It should be noted that for those skilled in the art, any improvements and modifications made without departing from the principles of the present invention should also be considered within the scope of protection of the present invention.
Claims
1. An automatic screening method for superior and diseased Camellia oleifera plants based on UAV imagery, characterized in that, Includes the following steps: S1, Data Acquisition and Stitching: Control a drone equipped with a multispectral camera, a visible light camera and a high-precision positioning module to take aerial photos of the target camellia oleifera forest, acquire multispectral images and visible light images with geographic coordinates, and stitch the multispectral images and visible light images to generate an orthophoto map with geographic reference information. S2, Canopy Extraction: Based on the orthophoto image obtained in step S1, multiple camellia tree canopies in the image are identified and located through a pre-trained camellia tree canopy target detection model, and the pixel position information of each canopy is obtained. The camellia tree canopy target detection model is a model based on a deep learning framework. S3, Health Assessment and Screening: Based on the multispectral image data acquired in S1, calculate the growth assessment index reflecting the physiological state of each canopy identified in S2; statistically sort the growth assessment index values of all canopies, and automatically screen out the canopy subsets characterized as superior and diseased plants according to preset rules. S4. Output Results: Using the georeferenced information included in the TIFF file, batch conversion from pixel coordinates to latitude and longitude is achieved, and the filtered results containing the geographical location information of the canopies of the superior and diseased trees are output. These results are then imported into the GIS system for precise management.
2. The method for automatic screening of superior and diseased Camellia oleifera plants based on UAV imagery according to claim 1, characterized in that, In step S1, the high-precision positioning module is a real-time dynamic differential positioning module or a precise single-point positioning module, used to enable aerial images to have centimeter-level or sub-meter-level geographic positioning accuracy.
3. The method for automatic screening of superior and diseased Camellia oleifera plants based on UAV imagery according to claim 1, characterized in that, In step S1, the multispectral camera covers at least two non-visible light bands, including the near-infrared band; the orthophoto is a GeoTIFF format file containing geographic coordinates and multiple spectral band information.
4. The method for automatic screening of superior and diseased Camellia oleifera plants based on UAV imagery according to claim 1, characterized in that, Step S2 specifically includes: S21, Image slicing: The orthophoto image or its derived visible light image is segmented to generate multiple slice images; S22, Model recognition: Input the sliced image into a pre-trained camellia tree canopy target detection model, and automatically output the pixel coordinates of the outer frame of each canopy; S23, Result Correction: Perform manual or automated post-processing on the results of automatic model recognition to delete or correct incorrectly identified non-canopy targets.
5. The method for automatic screening of superior and diseased Camellia oleifera plants based on UAV imagery according to claim 1, characterized in that, In step S3, the growth assessment index is the vegetation index, which is calculated based on the reflectance of the near-infrared band and the red band in the multispectral image.
6. The method for automatic screening of superior and diseased Camellia oleifera plants based on UAV imagery according to claim 1, characterized in that, In step S3, the multispectral image data is used to calculate the average vegetation index of each camellia tree canopy. The indices are sorted from smallest to largest, and the 3% of plants with the lowest indices are selected as diseased plants, while the 3% of plants with the highest indices are selected as superior plants. The outer frame pixel coordinates of the diseased plants and superior plants are output respectively. The average vegetation index includes GVI and NDVI, and its calculation formula is as follows: GVI = (NIR + Red) / (NIR - Red) NDVI = (NIR - Red) / (NIR + Red) NIR stands for near-infrared band, and Red stands for red band.
7. The method for automatic screening of superior and diseased Camellia oleifera plants based on UAV imagery according to claim 1, characterized in that, The growth assessment index may also be any one of the following: normalized difference vegetation index, greenness vegetation index, enhanced vegetation index, or soil-regulated vegetation index.
8. The method for automatic screening of superior and diseased Camellia oleifera plants based on UAV imagery according to claim 1, characterized in that, In step S3, the preset rule is as follows: the tree canopy that is in the highest value range in the growth assessment index ranking and accounts for a first predetermined proportion is judged as a superior tree; the tree canopy that is in the lowest value range and accounts for a second predetermined proportion is judged as a diseased tree, wherein the first predetermined proportion and the second predetermined proportion are equal or unequal.
9. The method for automatic screening of superior and diseased Camellia oleifera plants based on UAV imagery according to claim 1, characterized in that, Step S4 includes: S41, Coordinate Transformation: Based on the geographic reference information of the orthophoto map, the pixel coordinates of the center points of the crowns of the superior and diseased trees selected in step S3 are converted into latitude and longitude coordinates. S42, Formatted Output: Output the converted coordinate information, along with the tree crown identifier and growth assessment index value, as a structured data file or database record; The formula for calculating the pixel coordinates of the center point of the tree canopy outline is: X coordinate = (top left X + bottom right X) / 2, Y coordinate = (top left Y + bottom right Y) / 2.
10. An automatic screening system for superior and diseased Camellia oleifera plants based on UAV imagery, used to implement the automatic screening method for superior and diseased Camellia oleifera plants based on UAV imagery as described in any one of claims 1 to 9, characterized in that, include: The data acquisition unit includes an unmanned aerial vehicle (UAV) platform, as well as a multispectral imaging device, a visible light imaging device, and a high-precision positioning device mounted on it, for acquiring image data of camellia oleifera forests with geographic coordinates. A data processing and analysis unit, communicatively connected to the data acquisition unit, includes at least one processor and a memory. The memory stores a computer program, which, when executed by the processor, performs the functions of the following modules: The image processing module is used to stitch together the acquired image data to generate orthophoto maps; The canopy recognition module is used to automatically identify and locate the canopy of camellia oleifera trees from images; The health assessment module is used to calculate the growth assessment index of the tree canopy based on multispectral data, and automatically screen the superior trees and diseased trees according to the index ranking. The results generation module is used to convert the pixel coordinates of the filtered results into geographic coordinates and output them.