Recognition system for shooting corn seedlings based on unmanned aerial vehicle
Through the image processing system of field corn seedlings photographed by drone, high-resolution cameras and image processing technology are used to solve the problem of high image blur and overlap rate, improve the recognition accuracy and efficiency of corn seedlings, and form the optimal recognition model , realizes accurate identification of corn seedlings of different types and regions.
Patent Information
- Application Number
- CN202411228100.X
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-09-03
- Publication Date
- 2024-11-15
AI Technical Summary
During the process of taking corn seedling images, existing drones have caused blurred images, low recognition accuracy and efficiency due to weather conditions, shooting overlap rate and light conditions.
The image processing system for field corn seedlings is adopted based on drone shooting, including the drone image acquisition system, image processing system and data set training system. The images are collected through a 10x zoom ultra-definition camera and a 12-megapixel wide-angle camera, and cropped and grayscale The Brenner gradient function is used to evaluate clarity, perform data augmentation and model optimization, and form an optimal identification model.
It improves the accuracy and efficiency of image recognition, enhances the generalization and recognition ability of the model, and ensures the accuracy of the corn seedling recognition task under different conditions.
Smart Images

Figure CN118968366A_ABST
Abstract
Description
Technical Field
[0001] The invention relates to the field of image processing, and in particular to an image processing system for corn seedlings in a field photographed by an unmanned aerial vehicle. Background Art
[0002] Drone photography of seedlings is a common method for evaluating the growth status of seedlings and is widely used. However, in the process of existing drone photography of corn seedlings, images are blurred and there is little qualified image data, which reduces the accuracy and efficiency of corn seedling recognition. The present invention provides an image processing system for corn seedlings photographed in the field by drones. Summary of the invention
[0003] In view of the shortcomings of the prior art, the present invention provides an image processing system for field corn seedlings photographed by a drone, which solves the problem of low recognition accuracy and efficiency caused by blurred images and high shooting overlap rate due to the influence of weather conditions, shooting overlap rate and light conditions during the shooting process.
[0004] To achieve the above objectives, the present invention is implemented through the following technical solutions: a system for processing images of corn seedlings in the field shot by drones, comprising:
[0005] The drone image acquisition system collects image data of corn seedlings in the field, and includes a 10x zoom ultra-clear camera and a 12-megapixel wide-angle camera;
[0006] The image processing system receives and processes the image data taken by the drone, and includes: a data receiving module receives the image data taken by the drone and transmits it to the image processing module; the image processing module completes the selection and data enhancement of the corn seedling image data;
[0007] The data set training system makes the collected corn seedling image data into a data set, performs initial model training on the data set, and performs parameter adjustment and model optimization operations based on the initial model to form an optimal recognition model;
[0008] The corn seedling recognition system is configured with the optimal recognition model to complete the task of identifying corn seedlings of different types and regions.
[0009] Drone shooting method: The shooting height of the drone to obtain images is 15m, the shooting angle is 45° downward, the flight speed is 15km / h, and the shooting speed is 3s / frame.
[0010] The image processing module of the image processing system includes:
[0011] The specific steps of the image data screening unit include:
[0012] A1. First, crop the color image of corn seedlings into a pixel size of 640*640;
[0013] A2. Then, use the Luma formula f (x,y) =0.299R+0.587G+0.114B Convert the RGB value of each pixel of the corn seedling color image into a grayscale value and convert it into a grayscale image. (x,y) is the grayscale value, R, G, B are the red, green and blue components of the pixel respectively;
[0014] A3. Apply the Brenner gradient function to evaluate the clarity of the grayscale image. The calculation formula is as follows: H(f) = ∑ x,y (f (x+1,y) -f (x,y) ) 2 +(f (x,y+1) -f (x,y) ) 2 Among them, f (x,y) is the grayscale value of the grayscale image at the pixel (x, y), H (f) Indicates the result of image clarity calculation;
[0015] A4, Image clarity H (f) = N, it means the image is qualified. (f) ≥N, keep the corresponding color image. (f) When <N, delete the corresponding color picture;
[0016] The image enhancement unit performs data enhancement on the color image of corn seedlings. The enhancement methods include:
[0017] B1, image rotation, x1 = xcos(θ)-ysin(θ)+Δx, y1 = xsin(θ)+ycos(θ)+Δy, where (x1, y1) are the coordinates after rotation, θ is the rotation angle, and (Δx, Δy) is the translation vector used to keep the center of the image within the image after rotation;
[0018] B2. Horizontal mirroring of the image. Horizontal mirroring is to flip the image along the vertical center axis. The transformation formula is as follows: x2=x, y2=Dy, (x, y) is the point coordinates in the original image, (x2, y2) is the point coordinates after mirroring, and D is the height of the image.
[0019] The specific implementation of the recognition model building module is as follows:
[0020] S1. First, adjust the parameters of the initial model, adjust the number of training times, the number of images per iteration, and the learning rate;
[0021] S2, then, find the best parameter combination to optimize the model;
[0022] S3. Use untrained images to test the model and verify the feasibility of the optimal model.
[0023] The present invention provides an image processing system for corn seedlings in the field based on drone photography. Compared with the prior art, it has the following beneficial effects:
[0024] The image processing system based on the image processing of corn seedlings in the field taken by drones can crop the image data into a uniform size of 640mm*640mm through the image processing system, effectively reducing the overlap of image data, which can be better applied to model training and improve the generalization of the model.
[0025] The image processing system of corn seedlings in the field shot by drone can grayscale the image data and calculate the blur degree of the image data to identify unqualified images, thus improving the accuracy of model training.
[0026] The image processing system of corn seedlings in the field taken by drone completes the data set production of image data through the data training system, uses the initial model for training, learns the corn seedling recognition function, and analyzes the performance of the initial model. The parameters of the initial model are adjusted according to the performance, and the optimal parameter combination is found to achieve optimization and configure the optimal recognition model. The corn seedling recognition module is used to complete the recognition task of corn seedlings of different types and regions, thereby improving the recognition ability. BRIEF DESCRIPTION OF THE DRAWINGS
[0027] Figure 1 Workflow diagram of the image processing system for field corn seedlings photographed by drone. DETAILED DESCRIPTION
[0028] In view of the shortcomings of the prior art, the present invention provides an image processing system for field corn seedlings photographed by a drone, which solves the problem of low recognition accuracy and efficiency caused by blurred images and high shooting overlap rate due to the influence of weather conditions, shooting overlap rate and light conditions during the shooting process.
[0029] To achieve the above objectives, the present invention is implemented through the following technical solutions, based on the image processing system of corn seedlings photographed in the field by drones, including:
[0030] The drone image acquisition device collects image data of corn seedlings in the field, and includes a 10x zoom ultra-clear camera and a 12-megapixel wide-angle camera;
[0031] The image processing system receives and processes the image data taken by the drone, and includes: a data receiving module receives the image data taken by the drone and transmits it to the image processing module; the image processing module completes the selection and data enhancement of the corn seedling image data;
[0032] The data set training system makes the collected corn seedling image data into a data set, performs initial model training on the data set, and performs parameter adjustment and model optimization operations based on the initial model to form an optimal recognition model;
[0033] The corn seedling recognition system is configured with the optimal recognition model to complete the task of identifying corn seedlings of different types and regions.
[0034] Drone shooting method: The shooting height of the drone to obtain images is 15m, the shooting angle is 45° downward, the flight speed is 15km / h, and the shooting speed is 3s / frame.
[0035] The image processing module of the image processing system includes:
[0036] The specific steps of the image data screening unit include:
[0037] A1. First, crop the color image of corn seedlings into a pixel size of 640*640;
[0038] A2. Then, use the Luma formula f (x,y) =0.299R+0.587G+0.114B Convert the RGB value of each pixel of the corn seedling color image into a grayscale value and convert it into a grayscale image. (x,y) is the grayscale value, R, G, B are the red, green and blue components of the pixel respectively;
[0039] A3. Apply the Brenner gradient function to evaluate the clarity of the grayscale image. The calculation formula is as follows: H(f) = ∑ x,y (f (x+1,y) -f (x,y) ) 2 +(f (x,y+1) -f (x,y) ) 2 Among them, f (x,y) is the grayscale value of the grayscale image at the pixel (x, y), H (f) Indicates the result of image clarity calculation;
[0040] A4, Image clarity H (f) = N, it means the image is qualified. (f) ≥N, keep the corresponding color image. (f) When <N, delete the corresponding color picture;
[0041] The image enhancement unit performs data enhancement on the color image of corn seedlings. The enhancement methods include:
[0042] B1, image rotation, x1 = xcos(θ)-ysin(θ)+Δx, y1 = xsin(θ)+ycos(θ)+Δy, where (x1, y1) are the coordinates after rotation, θ is the rotation angle, and (Δx, Δy) is the translation vector used to keep the center of the image within the image after rotation;
[0043] B2. Horizontal mirroring of the image. Horizontal mirroring is to flip the image along the vertical center axis. The transformation formula is as follows: x2=x, y2=Dy, (x, y) is the point coordinates in the original image, (x2, y2) is the point coordinates after mirroring, and D is the height of the image.
[0044] The specific implementation of the data set production module is as follows:
[0045] C1. Use annotation tools to annotate the images and place the text files and image files of the dataset in different folders;
[0046] C2. Divide the data set into training set, validation set, and test set;
[0047] The specific implementation of the model training module is as follows:
[0048] D1. Perform initial model training on the data set;
[0049] D2. Analyze various indicators of the model and select the appropriate model;
[0050] The specific implementation of the recognition model building module is as follows:
[0051] S1. First, adjust the parameters of the initial model, adjust the number of training times, the number of images per iteration, and the learning rate;
[0052] S2, then, find the best parameter combination to optimize the model;
[0053] S3. Use untrained images to test the model and verify the feasibility of the optimal model.
[0054] Comparative test:
[0055] The images directly acquired by the drone image acquisition device at different periods and in different areas and the images processed by the drone-based image processing system of corn seedlings in the field were used to create data sets, train models, and optimize models. The same data was used for model testing, and the comparison results are shown in the following table.
[0056]
[0057] In summary, after processing the images of the four areas A, B, C, and D by the image processing system based on drone-photographed field corn seedlings provided by the present invention, the recognition accuracy is greatly improved, indicating that the images processed by the image processing system of the present invention can improve the accuracy of the model, which is of positive significance for promoting more accurate identification of corn seedlings.
Claims
1. Based on the image processing system of corn seedlings photographed in the field by drone, it is characterized by: include: The drone image acquisition system collects image data of corn seedlings in the field, and includes a 10x zoom ultra-clear camera and a 12-megapixel wide-angle camera; The image processing system receives and processes the image data taken by the drone, and includes: a data receiving module receives the image data taken by the drone and transmits it to the image processing module; the image processing module completes the selection and data enhancement of the corn seedling image data; The data set training system makes the collected corn seedling image data into a data set, performs initial model training on the data set, and performs parameter adjustment and model optimization operations based on the initial model to form an optimal recognition model; The corn seedling recognition system is configured with the optimal recognition model to complete the task of identifying corn seedlings of different types and regions.
2. According to claim 1, the image processing system for corn seedlings in the field based on drone photography is characterized by: The drone acquires images at a shooting height of 15m, a shooting angle of 45° downward, a flight speed of 15km / h, and a shooting speed of 3s / frame.
3. The image processing system for corn seedlings in the field based on drone photography according to claim 1, characterized in that: The image processing module of the image processing system includes: The specific steps of the image data screening unit include: A1. First, crop the color image of corn seedlings into a pixel size of 640*640; A2. Then, use the Luma formula f (x,y) =0.299R+0.587G+0.114B Convert the RGB value of each pixel of the corn seedling color image into a grayscale value and convert it into a grayscale image. (x,y) is the grayscale value, R, G, B are the red, green and blue components of the pixel respectively; A3. Apply the Brenner gradient function to evaluate the clarity of the grayscale image. The calculation formula is as follows: H(f) = ∑ x,y (f (x+1,y) -f (x,y) ) 2 +(f (x,y+1) -f (x,y) ) 2 Among them, f (x,y) is the grayscale value of the grayscale image at the pixel (x, y), H (f) Indicates the result of image clarity calculation; A4, Image clarity H (f) = N, it means the image is qualified. (f) ≥N, keep the corresponding color image. (f) When <N, delete the corresponding color picture; The image enhancement unit performs data enhancement on the color image of corn seedlings. The enhancement methods include: B1, image rotation, x1 = xcos(θ)-ysin(θ)+Δx, y1 = xsin(θ)+ycos(θ)+Δy, where (x1, y1) are the coordinates after rotation, θ is the rotation angle, and (Δx, Δy) is the translation vector used to keep the center of the image within the image after rotation; B2. Horizontal mirroring of the image. Horizontal mirroring is to flip the image along the vertical center axis. The transformation formula is as follows: x2=x, y2=Dy, (x, y) is the point coordinates in the original image, (x2, y2) is the point coordinates after mirroring, and D is the height of the image.
4. The image processing system for corn seedlings in the field based on drone photography according to claim 1, characterized in that: The specific implementation of the recognition model building module is: S1. First, adjust the parameters of the initial model, adjust the number of training times, the number of images per iteration, and the learning rate; S2, then, find the best parameter combination to optimize the model; S3. Use untrained images to test the model and verify the feasibility of the optimal model.
Citation Information
Patent Citations
Sugarcane seedling automatic identification and counting method based on deep neural network
CN113392775A
Method, system and equipment for evaluating corn emergence rate based on deep learning
CN116721366A
