A method for quickly obtaining abnormal grass sample area based on drone aerial photography

Abstract

The present invention proposes a method for quickly obtaining the area of ​​abnormal grass samples based on unmanned aerial vehicle aerial photography, which belongs to the fields of remote sensing technology, deep learning technology and area calculation. The present invention uses an unmanned aerial vehicle to quickly and automatically obtain the number of pixels of target abnormal grass samples in each rectangular area and the total number of pixels in each rectangular area, and brings the obtained pixel parameters into a calculation formula for the area of ​​abnormal grass samples in the entire target area, so as to quickly obtain the area of ​​abnormal grass samples in the target area. The unmanned aerial vehicle of the method shoots through the shortest route, which is energy-saving and efficient. Compared with calculating the proportion of target abnormal grass sample pixels in each area once, the method for obtaining the total area calculation method of target abnormal grass samples in the entire target area will be simpler. During the shooting process of the unmanned aerial vehicle, the pixels of the rectangular area image shot previously can be counted. When the end of the route is reached, all pixel parameters are brought into the formula to quickly obtain the area of ​​abnormal grass samples in the entire target area.

Description

Technical Field

[0001] The present invention belongs to the fields of remote sensing technology, deep learning technology and area calculation, and specifically, relates to a method for quickly obtaining abnormal grass sample area based on drone aerial photography. Background Art

[0002] Abnormal weeds refer to the phenomenon that alien plant organisms are introduced into new ecosystems, grow and reproduce in that environment, and have a significant impact on the local ecosystem; alien species usually lack natural enemies in their new environment, so they can spread rapidly, leading to the reduction or extinction of native species, disrupting the ecological balance, and may have a negative impact on human society and economy.

[0003] For example, Spartina alterniflora is a typical abnormal grass species. It has strong growth and spread capabilities and can quickly occupy large areas of wetlands. The hazards caused by Spartina alterniflora include ecological damage, loss of habitats for other organisms, and impacts on fisheries.

[0004] However, in the weed control of abnormal weeds, the difficulties encountered, in addition to biological characteristics, unsuitable control methods and resource limitations, are mainly due to the difficulty in monitoring and evaluating abnormal weeds. Among them, the assessment of the area of ​​abnormal weeds is crucial because it directly reflects the invasion status of abnormal weeds in the target area. At the same time, obtaining the invasion status and location of abnormal weeds in the target area can formulate targeted measures to control abnormal weeds in the area. Summary of the invention

[0005] The purpose of the present invention is to overcome the above problems existing in the prior art and greatly improve its technical effect on the basis of the original technology; to this end, the present invention provides a method for quickly obtaining abnormal grass sample area based on drone aerial photography, the method comprising:

[0006] (1) Determine the target area to be monitored, grid the target area, and divide the target area into multiple rectangular areas of equal area;

[0007] (ii) Determine the coordinate position of the center point of each rectangular area through the GPS application, and determine the shortest route for the drone to take images based on the coordinate position of the center point of each rectangular area; and determine the altitude H of the drone based on the area S1 of the rectangle, the photosensitive area S2 of the image sensor of the drone camera module, and the focal length F of the lens of the drone camera module. The formula for determining the altitude of the drone is:

[0008]

[0009] (3) Perform drone flight photography based on the determined shortest route and altitude H, perform vertical aerial photography at the coordinate position of the center point of each rectangular area, and obtain vertical aerial images of the corresponding rectangular area;

[0010] (iv) segmenting and processing the vertical aerial images of each rectangular area by using CNN deep learning technology to obtain segmented images, and calculating and identifying the number of pixels of the target abnormal grass sample in each rectangular area and the total number of pixels in each rectangular area by using the segmented images;

[0011] (V) Obtaining the area of ​​the abnormal grass sample in the entire target area according to the number of pixels of the abnormal grass sample in each rectangular area and the total number of pixels in each rectangular area; the calculation formula for the area of ​​the abnormal grass sample in the entire target area is:

[0012]

[0013] Among them, S represents the area of ​​abnormal grass samples in the entire target area, i represents the i-th rectangular area, n represents that the target area is divided into n rectangular areas through gridding, and α i It represents the number of pixels of the target abnormal grass sample in the i-th rectangular area, S1 represents the area of ​​each identical rectangle, and β0 represents the total number of pixels in each rectangular area image.

[0014] The target area is divided into a plurality of rectangular areas of equal area, including: the target area refers to an area to be monitored where abnormal target grass samples grow, the target area is gridded by TileMill software, and the target area is divided into a plurality of rectangles of equal length and width, and the lengths of all rectangles are parallel to each other, and the widths are parallel to each other; if the edge of the target area cannot be directly divided into rectangles, the target area is divided by including the edge area in rectangles of equal area.

[0015] The method of determining the shortest route when the drone takes images based on the coordinate positions of the center points of each rectangular area includes: the drone takes pictures from the center point of one rectangular area and then flies to the center point of an adjacent rectangular area to take pictures. The method of determining the shortest route when the drone takes images is: (a) determining the coordinate positions of the center points of each rectangle through a GPS application, and forming a scatter plot through the coordinate positions of all the center points; (b) selecting any point in the scatter plot as the starting point, and obtaining the shortest distance from the starting point to the coordinate positions of all other center points through the relevant algorithms of the open traveling salesman problem, and the shortest distance is the shortest route when the drone takes images; the open traveling salesman problem means that the goal is to visit all given locations, and to reach the end point from the starting point without returning to the starting point; the relevant algorithms for solving the open traveling salesman problem include: exact algorithms, ready-made tools and libraries, heuristics and approximate algorithms.

[0016] The formula for determining the altitude of the drone includes: the aspect ratios of S1 and S2 in the formula are the same, and in order to obtain clearer abnormal grass sample images and to ensure that the task volume is not too large, the size of S1 should be moderate; the size of S1 is determined based on the category characteristics of the target abnormal grass sample, the recognition effect of the abnormal grass sample in the captured image, and the workload.

[0017] The process of obtaining the vertical aerial image of the corresponding rectangular area includes: if the target area to be monitored is a multi-steep slope area, the obtained vertical aerial image needs to be orthorectified to obtain the corresponding orthorectified image; then, the orthorectified image is processed in steps (four) and (five) to obtain the area of ​​abnormal grass sample in the target area.

[0018] The CNN deep learning technology includes: CNN deep learning technology is a deep learning model for processing and analyzing image data; CNN automatically extracts image features through a series of convolutional layers, pooling layers and fully connected layers; CNN deep learning technology includes: model training, model verification and trained model; wherein, model training includes: first, collecting plant image data containing target abnormal grass samples, using annotation tools to annotate the plants in each image, and classifying them into corresponding labels; performing image scaling, normalization and data enhancement preprocessing tasks on the annotated images, and dividing the preprocessed images into training sets and test sets; secondly, selecting a suitable CNN architecture to form a CNN network model, and commonly used CNN architectures include VGG16, ResNet and Inception; finally, inputting the training set into the selected CNN network model for training until the set fixed number of rounds is reached and the loss function converges, then stopping the training; the model verification includes: inputting the test set into the trained model, and evaluating the performance of the model according to the results; according to the evaluation results, adjusting the hyperparameters of the model to improve the model performance until a model with good performance is obtained, and the adjusted model is called a trained model.

[0019] The CNN deep learning technology also includes: segmenting and processing the vertical aerial image of each rectangular area through a trained model to obtain a segmented image; the process of segmenting the image is: feature extraction, context capture and pixel classification; through the obtained segmented image, binary mask processing is performed on the segmented image to obtain a binary mask image of the target abnormal grass sample, wherein each pixel value of the target abnormal grass sample corresponding area is 1, and the background pixel value is 0; the number of pixels in the target abnormal grass sample area is obtained through the binary mask image; and the total number of pixels of the entire rectangular image is obtained through the Python image processing tool.

[0020] The method of obtaining the area of ​​the abnormal grass sample in the entire target area according to the number of pixels of the abnormal grass sample in each rectangular area and the total number of pixels in each rectangular area comprises: setting a central control center connected to the drone, and the central control center is equipped with TileMill software, GPS application, CNN deep learning technology, Python image processing tools and multiple algorithms; when the drone automatically performs vertical aerial photography according to the shortest route and altitude, the central control center immediately calculates the number of pixels of the abnormal grass sample in the image for each image obtained, and calculates the total number of pixels of each rectangular image; until the drone navigates to the end of the route, and calculates the number of pixels of the abnormal grass sample in the rectangle corresponding to the end of the route and the total number of pixels in each rectangular image, and brings the obtained number of pixels of the abnormal grass sample in the rectangle and the total number of pixels in each rectangular image into the calculation formula of the abnormal grass sample area of ​​the entire target area in step (five), so as to quickly obtain the area of ​​the abnormal grass sample in the entire target area.

[0021] The beneficial effects of the present invention are:

[0022] (1) The present invention provides a method for quickly obtaining abnormal grass sample areas based on drone aerial photography. The method can quickly obtain abnormal grass sample areas in the target area, providing important data support for formulating further weeding plans; and provides a method for obtaining the shortest flight path of a drone, so that the drone can take pictures through the shortest flight path, which is energy-saving and efficient.

[0023] (2) The present invention divides the target area into the same rectangular areas in a gridding manner, and directly calculates the total area of ​​the target abnormal grass samples in the entire target area by calculating the number of pixels of the target abnormal grass samples in each rectangular area image and the total number of pixels in each rectangular area image. Because the target abnormal grass samples photographed in each rectangular area will be clearer, the area calculation will be more accurate compared to directly obtaining the image of the entire target area. And because the area of ​​each rectangle is equal and the total pixels of the rectangular area image are equal, compared to calculating the proportion of the target abnormal grass sample pixels in each area once, the method for calculating the total area of ​​the target abnormal grass samples in the entire target area of ​​the present invention will be simpler.

[0024] (3) During the drone shooting process, the present invention can count the number of target abnormal grass-like pixels and the total number of pixels in the rectangular area image taken on the shortest route before; when the route end point is reached and the number of target abnormal grass-like pixels and the total number of pixels in the rectangular area image corresponding to the end point are calculated, the pixel parameters obtained in the entire route are directly substituted into the calculation formula of the abnormal grass-like area of ​​the entire target area, so as to quickly obtain the abnormal grass-like area of ​​the entire target area. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] Figure 1 The present invention is a flow chart of a method for quickly acquiring abnormal grass sample area based on drone aerial photography. DETAILED DESCRIPTION

[0026] The specific embodiments of the present invention are described in detail below in conjunction with the accompanying drawings; it should be understood that the specific embodiments given here are only used to illustrate and explain the present invention and cannot be used to limit the present invention.

[0027] like Figure 1 As shown, it is a flow chart of a method for quickly obtaining an abnormal grass sample area based on drone aerial photography of the present invention, and the flow chart includes: step S100, determining a target area to be monitored, gridding the target area, and dividing the target area into a plurality of rectangular areas of equal area; step S101, determining the coordinate position of the center point of each rectangular area through a GPS application, and determining the shortest route when the drone takes an image according to the coordinate position of the center point of each rectangular area; and determining the flight altitude H of the drone according to the area S1 of the rectangle, the photosensitive area S2 of the image sensor of the drone camera module, and the focal length F of the lens of the drone camera module; Step S102, perform drone flight photography according to the determined shortest route and flight altitude H, perform vertical aerial photography at the coordinate position of the center point of each rectangular area, and obtain the vertical aerial image of the corresponding rectangular area; Step S103, segment and process the vertical aerial image of each rectangular area through CNN deep learning technology, obtain the segmented image, and calculate and identify the number of pixels of the target abnormal grass sample in each rectangular area and the total number of pixels in each rectangular area through the segmented image; Step S104, obtain the area of ​​the abnormal grass sample in the entire target area according to the number of pixels of the target abnormal grass sample in each rectangular area and the total number of pixels in each rectangular area.

[0028] Specifically, the present invention first determines the target area to be monitored, grids the target area, and divides the target area into a plurality of rectangular areas of equal area; determines the coordinate position of the center point of each rectangular area through a GPS application, and determines the shortest route for a drone to shoot images according to the coordinate position of the center point of each rectangular area; and determines the altitude H of the drone through a series of parameters; then, the drone is flown according to the determined shortest route and the altitude H, and vertical aerial photography is performed at the coordinate position of the center point of each rectangular area to obtain a vertical aerial image of the corresponding rectangular area; and the vertical aerial image of each rectangular area is segmented and processed by a CNN deep learning technology to obtain a segmented image, and the number of pixels of the target abnormal grass sample of each rectangular area and the total number of pixels of each rectangular area are calculated and identified through the segmented image; finally, the area of ​​the abnormal grass sample of the entire target area is obtained according to the number of pixels of the target abnormal grass sample of each rectangular area and the total number of pixels of each rectangular area.

[0029] Step S100, determining a target area to be monitored, gridding the target area, and dividing the target area into a plurality of rectangular areas of equal area.

[0030] In the above embodiment, specifically, the target area refers to the area to be monitored where target abnormal grass samples grow. The target area is gridded using TileMill software to divide the target area into multiple rectangles with equal length and width, and the lengths of all rectangles are parallel to each other, and the widths are parallel to each other. In the case where the edge of the target area cannot be directly divided into rectangles, the edge area is divided by including rectangles of equal area in them, which facilitates the subsequent calculation of the total area of ​​the target area.

[0031] Step S101, determine the coordinate position of the center point of each rectangular area through the GPS application, and determine the shortest route for the drone to take images based on the coordinate position of the center point of each rectangular area; and determine the altitude H of the drone based on the area S1 of the rectangle, the photosensitive area S2 of the image sensor of the drone camera module, and the focal length F of the lens of the drone camera module.

[0032] Specifically, the formula for determining the altitude of a drone is:

[0033]

[0034] In the above embodiment, specifically, the aspect ratios of S1 and S2 in the formula are the same. In order to obtain a clearer abnormal grass sample image and to ensure that the task volume is not too large, the size of S1 should be moderate; the size of S1 is determined based on the category characteristics of the target abnormal grass sample, the recognition effect of the abnormal grass sample in the captured image, and the workload.

[0035] In the above embodiment, for example, the image sensor size of the DJI Zenmuse X7 is 22.5x16.7mm, the lens focal length is 24mm, and the actual shooting area that the camera wants to identify the flower and grass category under this lens is 16.3m 2 , that is, the length and width of the rectangle are 4.69m and 3.48m respectively; substituting all the data into the formula of the drone's altitude, the altitude H is 5m.

[0036] In the above embodiment, specifically, the drone takes a picture from the center point of a rectangular area and then flies to the center point of an adjacent rectangular area to take pictures. The method for determining the shortest route when the drone takes pictures is: (a) determining the coordinate positions of the center points of each rectangle through a GPS application, and forming a scatter plot through the coordinate positions of all the center points; (b) selecting any point in the scatter plot as the starting point, and obtaining the shortest distance from the starting point to the coordinate positions of all other center points through the relevant algorithms of the open traveling salesman problem, and the shortest distance is the shortest route when the drone takes pictures; the open traveling salesman problem means that the goal is to visit all given locations and reach the end point from the starting point without returning to the starting point; the relevant algorithms for solving the open traveling salesman problem include: exact algorithms, ready-made tools and libraries, heuristics and approximate algorithms.

[0037] In the above embodiment, specifically, when calculating the shortest distance of drone navigation through the open traveling salesman problem related algorithm, the starting point is any edge point of the scatter plot to obtain the shortest route when the drone takes the image.

[0038] Step S102 , performing drone flight photography according to the determined shortest route and flight altitude H, performing vertical aerial photography at the coordinate position of the center point of each rectangular area, and obtaining a vertical aerial photography image of the corresponding rectangular area.

[0039] In the above embodiment, specifically, if the target area to be monitored is an area with multiple steep slopes, it is necessary to perform orthorectification on the obtained vertical aerial image to obtain the corresponding orthophoto image; then, the orthophoto image is processed in steps S103 and S104 to obtain the area of ​​abnormal grass samples in the target area.

[0040] Step S103, segment and process the vertical aerial image of each rectangular area through CNN deep learning technology to obtain a segmented image, and calculate and identify the number of pixels of abnormal grass samples in each rectangular area and the total number of pixels in each rectangular area through the segmented image.

[0041] In the above embodiment, specifically, CNN deep learning technology is a deep learning model for processing and analyzing image data; CNN automatically extracts image features through a series of convolutional layers, pooling layers and fully connected layers; CNN deep learning technology includes: model training, model verification and trained model; wherein, model training includes: first, collecting plant image data containing target abnormal grass samples, using annotation tools to annotate the plants in each image, and classifying them into corresponding labels; performing image scaling, normalization and data enhancement preprocessing tasks on the annotated images, and dividing the preprocessed images into training sets and test sets; secondly, selecting a suitable CNN architecture to form a CNN network model, and commonly used CNN architectures include VGG16, ResNet and Inception; finally, inputting the training set into the selected CNN network model for training until the set fixed number of rounds is reached and the loss function converges, etc., then the training is stopped; the verification of the model includes: inputting the test set into the trained model, and evaluating the performance of the model according to the results; according to the evaluation results, adjusting the hyperparameters of the model to improve the model performance until a model with good performance is obtained, and the adjusted model is called a trained model.

[0042] In the above embodiment, specifically, the vertical aerial image of each rectangular area is segmented and processed through the trained model to obtain a segmented image; the process of segmenting the image is: feature extraction, context capture and pixel classification; through the obtained segmented image, the segmented image is binary masked to obtain a binary mask image of the target abnormal grass sample, wherein each pixel value of the corresponding area of ​​the target abnormal grass sample is 1, and the background pixel value is 0, and the number of pixels in the target area is obtained through the binary mask image; and the total number of pixels of the entire rectangular image is obtained through the Python image processing tool.

[0043] Step S104, obtaining the area of ​​the abnormal grass pattern in the entire target area according to the number of pixels of the abnormal grass pattern in each rectangular area and the total number of pixels in each rectangular area.

[0044] In the above embodiment, specifically, the area calculation formula of the abnormal grass pattern in the entire target area is:

[0045]

[0046] Among them, S represents the area of ​​abnormal grass samples in the entire target area, i represents the i-th rectangular area, n represents that the target area is divided into n rectangular areas through gridding, and α i Represents the number of pixels of the target abnormal grass sample in the i-th rectangular area, S1 represents the area of ​​each identical rectangle, β0 represents the total number of pixels in each rectangular area image, and β0 indicates that the total number of pixels in each rectangular area image is the same.

[0047] In the above embodiment, specifically, a central control center connected to the drone is set up, and the central control center is equipped with TileMill software, GPS application, CNN deep learning technology, Python image processing tools and various algorithms, etc.; the various algorithms include: grid computing algorithm, GPS positioning algorithm, deep learning algorithm, Python image processing algorithm, drone parameter conversion algorithm and various area calculation algorithms, etc.; when the drone automatically performs vertical aerial photography according to the shortest route and altitude, the drone will wirelessly transmit the image to the central control center for each image obtained, and the central control center immediately calculates the number of pixels of the target abnormal grass sample in the image and the total number of pixels of the corresponding rectangular area image until the drone navigates to the end of the route, and calculates the number of pixels of the target abnormal grass sample in the rectangle corresponding to the end of the route and the total number of pixels of the corresponding rectangular area image, and brings all the obtained pixel parameters into the calculation formula of the abnormal grass sample area of ​​the entire target area to quickly obtain the area of ​​the abnormal grass sample of the entire target area.

[0048] In the above embodiment, preferably, a GPS positioning device is installed on the drone, and the shortest route, altitude and intelligent control of the drone are set to automatically take fixed-point photos when the drone reaches the central area of ​​each rectangle, so that the drone can automatically complete the shooting of the entire target area after taking off from the initial position, and transmit the captured image information to the central control center, which calculates the area of ​​abnormal grass samples in the target area. This process can be fully automated, and the area of ​​abnormal grass samples in the target area can be quickly obtained, and the distribution of abnormal grass samples in the target area can be obtained, providing data support for the subsequent cleaning of abnormal grass samples.

[0049] It should be understood that the above-mentioned embodiments are one or more embodiments of the present invention, and there are many other embodiments and variations thereof based on the present invention; the variations and modifications made by ordinary technicians in this industry through the present invention without making pioneering innovations all fall within the scope of protection of the present invention.

Claims

1. A method for quickly obtaining abnormal grass sample area based on drone aerial photography, characterized in that: The method comprises: S100, determining a target area to be monitored, gridding the target area, and dividing the target area into a plurality of rectangular areas of equal area; S200, determine the coordinate position of the center point of each rectangular area through the GPS application, and determine the shortest route for the drone to shoot images based on the coordinate position of the center point of each rectangular area; and determine the altitude H of the drone based on the area S1 of the rectangle, the photosensitive area S2 of the image sensor of the drone camera module, and the focal length F of the lens of the drone camera module. The formula for determining the altitude of the drone is: S300, performing drone flight photography according to the determined shortest route and flight altitude H, performing vertical aerial photography at the coordinate position of the center point of each rectangular area, and obtaining a vertical aerial photography image of the corresponding rectangular area; S400, segmenting and processing the vertical aerial image of each rectangular area by using CNN deep learning technology to obtain a segmented image, and calculating and identifying the number of pixels of the abnormal grass sample of each rectangular area and the total number of pixels of each rectangular area by using the segmented image; S500, obtaining the area of ​​the abnormal grass sample in the entire target area according to the number of pixels of the abnormal grass sample in each rectangular area and the total number of pixels in each rectangular area; the calculation formula for the area of ​​the abnormal grass sample in the entire target area is: Among them, S represents the area of ​​abnormal grass samples in the entire target area, i represents the i-th rectangular area, n represents that the target area is divided into n rectangular areas through gridding, and α i It represents the number of pixels of the target abnormal grass sample in the i-th rectangular area, S1 represents the area of ​​each identical rectangle, and β0 represents the total number of pixels in each rectangular area image.

2. The method for quickly obtaining abnormal grass sample area based on drone aerial photography according to claim 1 is characterized in that: The target area is divided into a plurality of rectangular areas of equal area, including: the target area refers to the area to be monitored where abnormal target grass samples grow, the target area is gridded by TileMill software, and the target area is divided into a plurality of rectangles of equal length and width, and the lengths of all rectangles are parallel to each other, and the widths are parallel to each other; if the edge of the target area cannot be directly divided into rectangles, the target area is divided by including the edge area in rectangles of equal area.

3. The method for quickly obtaining abnormal grass sample area based on drone aerial photography according to claim 1 is characterized in that: The method of determining the shortest route when the drone takes images based on the coordinate positions of the center points of each rectangular area includes: a method of determining the shortest route when the drone takes images is: S201, determining the coordinate positions of the center points of each rectangle through a GPS application, and forming a scatter plot through the coordinate positions of all the center points; S202, selecting any point in the scatter plot as a starting point, and obtaining the shortest distance from the starting point to the coordinate positions of all other center points through a related algorithm of the open traveling salesman problem, wherein the shortest distance is the shortest route when the drone takes images; the goal of the open traveling salesman problem is to visit all given locations, and to reach the end point from the starting point without returning to the starting point; related algorithms for solving the open traveling salesman problem include: exact algorithms, heuristic algorithms, and approximate algorithms.

4. The method for quickly obtaining abnormal grass sample area based on drone aerial photography according to claim 1 is characterized in that: The formula for determining the altitude of the drone includes: the length-to-width ratio of the rectangles corresponding to S1 and S2 in the formula is the same, and in order to obtain a clearer abnormal grass sample image and to ensure that the task volume is not too large, the size of S1 should be moderate; the size of S1 is determined based on the category characteristics of the target abnormal grass sample, the recognition effect of the abnormal grass sample in the captured image, and the workload.

5. The method for quickly obtaining abnormal grass sample area based on drone aerial photography according to claim 1 is characterized in that: The process of obtaining the vertical aerial image corresponding to the rectangular area includes: if the target area to be monitored is a multi-steep slope area, the obtained vertical aerial image needs to be orthorectified to obtain the corresponding orthoimage; then, the orthoimage is processed in steps S400 and S500 to obtain the area of ​​abnormal grass sample in the target area.

6. The method for quickly obtaining abnormal grass sample area based on drone aerial photography according to claim 1 is characterized in that: The CNN deep learning technology includes: CNN deep learning technology is a deep learning model for processing and analyzing image data; CNN automatically extracts image features through a series of convolutional layers, pooling layers and fully connected layers; CNN deep learning technology includes: model training, model verification and trained model; wherein, model training includes: first, collecting plant image data containing target abnormal grass samples, using annotation tools to annotate the plants in each image, and classifying them into corresponding labels; performing image scaling, normalization and data enhancement preprocessing tasks on the annotated images, and dividing the preprocessed images into training sets and test sets; secondly, selecting a suitable CNN architecture to form a CNN network model, the CNN architecture includes VGG16, ResNet and Inception; finally, inputting the training set into the selected CNN network model for training until the set fixed number of rounds is reached and the loss function converges, then stopping the training; the model verification includes: inputting the test set into the trained model, and evaluating the performance of the model according to the results; according to the evaluation results, adjusting the hyperparameters of the model to improve the model performance until a model with good performance is obtained, and the adjusted model is called a trained model.

7. The method for quickly obtaining abnormal grass sample area based on drone aerial photography according to claim 1 is characterized in that: The CNN deep learning technology also includes: segmenting and processing the vertical aerial image of each rectangular area through a trained model to obtain a segmented image; the process of segmenting the image is: feature extraction, context capture and pixel classification; through the obtained segmented image, binary mask processing is performed on the segmented image to obtain a binary mask image of the target abnormal grass sample, wherein each pixel value of the target abnormal grass sample corresponding area is 1, and the background pixel value is 0; the number of pixels in the target abnormal grass sample area is obtained through the binary mask image; and the total number of pixels of the entire rectangular image is obtained through the Python image processing tool.

8. The method for quickly obtaining abnormal grass sample area based on drone aerial photography according to claim 1 is characterized in that: The method of obtaining the area of ​​the abnormal grass sample in the entire target area according to the number of pixels of the target abnormal grass sample in each rectangular area and the total number of pixels in each rectangular area includes: setting a central control center connected to the drone, and the central control center is equipped with TileMill software, GPS application, CNN deep learning technology and Python image processing tools; when the drone automatically performs vertical aerial photography according to the shortest route and altitude, the central control center immediately calculates the number of pixels of the target abnormal grass sample in the image for each image obtained, and calculates the total number of pixels of each rectangular image; until the drone navigates to the end of the route, and calculates the number of pixels of the target abnormal grass sample in the rectangle corresponding to the end of the route and the total number of pixels of each rectangular image, and brings the obtained number of pixels of the target abnormal grass sample in the rectangle and the total number of pixels of each rectangular image into the calculation formula of the area of ​​the abnormal grass sample in the entire target area in step S500, so as to quickly obtain the area of ​​the abnormal grass sample in the entire target area.

Images