Method, system, medium and equipment for acquiring local characters of corn

Abstract

The invention discloses a corn local character acquisition method, system, medium and equipment, and relates to the technical field of image recognition, and the method comprises the steps: inputting a corn image data set into a recognition model based on a neural network, training the recognition model, and obtaining a corn plant preliminary recognition model; obtaining a corn image of a to-be-detected area, inputting the corn image into the corn plant preliminary identification model, and performing fine tuning training on the model to obtain a plant frame identification model; inputting a to-be-identified corn image into the plant frame identification model, obtaining bounding box information of a corn plant, tracking and marking the corn plant, identifying a target part image of the corn plant, and determining a to-be-identified corn local character data set; constructing a local shape identification model; and inputting a to-be-recognized corn local character data set into the trained local character recognition model to obtain a final corn local character recognition result.

Description

Technical Field

[0001] The present invention relates to the field of image recognition, and in particular to a method, system, medium and equipment for obtaining local traits of corn. Background Art

[0002] In recent years, artificial intelligence and computer vision technologies have been gradually applied in agriculture, especially convolutional neural network (CNN)-based object detection algorithms (such as the YOLO series). These technologies can automatically identify and extract features of crop targets.

[0003] However, some existing automated equipment usually only supports simple parameter recording, such as plant height, but it is difficult to accurately identify complex biological traits such as the number of plants and the number of aerial root layers. At the same time, factors such as field lighting changes, plant occlusion, and differences in different growth stages limit the application of existing target detection technology in field scenes. It mainly focuses on the detection of macro traits (such as the number and position of plants), but has insufficient recognition capabilities for micro traits (such as the number of aerial root layers of corn). Especially in complex environments such as lighting and occlusion, the recognition effect is even worse. Summary of the Invention

[0004] The present invention provides a method, system, medium, and device for obtaining local traits of corn, which are used to solve the above-mentioned problem existing in the prior art, namely, how to accurately identify target traits of corn in the prior art. The present invention provides a method for obtaining local traits of corn, which includes: Obtain historical field corn plant images and construct a corn image dataset; Inputting the corn image dataset into a neural network-based recognition model, training the recognition model, and obtaining a preliminary corn plant recognition model for identifying corn characteristics; Obtain corn images from multiple test areas to construct a fine-tuning dataset, fine-tune the preliminary corn plant recognition model, and obtain a plant bounding box recognition model; The corn images to be identified on both sides of the test area collected in real time by the field robot are input into the plant border recognition model to obtain the bounding box information of each corn plant in each frame of corn image; Based on the bounding box information of each corn plant, each corn plant is tracked and independently marked, and the target part image of the marked corn plant is identified to determine the corn local trait dataset to be identified; A local trait recognition model based on a neural network was constructed. The corn local trait dataset to be identified was input into the trained local trait recognition model to determine multiple local trait recognition results of the corn plant with the same mark in each frame, and the local trait recognition result with the highest confidence was selected as the final corn local trait recognition result.

[0005] Optionally, tracking the trajectory of each corn plant and setting an independent mark for each corn plant according to the bounding box information of each corn plant specifically includes: The bounding box information of each corn plant is input into the BoT-SORT algorithm and ByteTrack algorithm to track the trajectory of the corn plant and obtain the corresponding trajectory information; According to the trajectory information, the double-line collision method is used to determine whether the area to which the corn plant belongs is the test area where the field robot is located. If so, a unique label is assigned to each corn plant in the current area.

[0006] Optionally, the YOLOv10 algorithm is used to identify the target part image of the marked corn plant.

[0007] Optionally, the neural network is a YOLOV10 network.

[0008] Optionally, acquiring historical field corn plant images and constructing a corn image dataset specifically includes: The historical field corn plant images were extracted at intervals of 10 frames, and 3-4 corn plants on both sides of the ridge were manually marked to construct a corn image dataset.

[0009] Optionally, an image enhancement operation is performed on the corn image dataset.

[0010] The present invention provides a system for obtaining local traits of corn, comprising: The acquisition module is used to obtain historical field corn plant images and construct a corn image dataset; A corn plant preliminary recognition model determination module is used to input the corn image data set into a recognition model based on a neural network, train the recognition model, and obtain a corn plant preliminary recognition model for identifying corn characteristics; The plant border recognition model determination module is used to obtain corn images of multiple test areas to construct a fine-tuning dataset, fine-tune the preliminary corn plant recognition model, and obtain a plant border recognition model; The bounding box information determination module is used to input the corn images to be identified on both sides of the test area collected in real time by the field robot into the plant border recognition model to obtain the bounding box information of each corn plant in each frame of corn image; The marking module is used to track the trajectory of each corn plant and set an independent mark based on the bounding box information of each corn plant, identify the target part image of the marked corn plant, and determine the corn local trait dataset to be identified; The recognition module is used to build a local trait recognition model based on a neural network; the corn local trait dataset to be identified is input into the trained local trait recognition model, multiple local trait recognition results of the corn plant with the same mark in each frame are determined, and the local trait recognition result with the highest confidence is selected as the final corn local trait recognition result.

[0011] The present invention provides a computer-readable storage medium, wherein the storage medium stores a computer program, and when the computer program is executed by a processor, the method for obtaining local traits of corn is implemented.

[0012] The present invention provides a computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, the above-mentioned method for obtaining local traits of corn is implemented.

[0013] Compared with the prior art, the beneficial effects of the present invention are as follows: the present invention provides a method for obtaining local traits of corn, which trains a recognition model based on a neural network through an acquired corn image data set to obtain a preliminary recognition model of corn plants, and then fine-tunes the model through a small amount of field data slices to obtain a plant border recognition model, which can accurately obtain the boundary box information of the corn plant and improve the accuracy of corn trait recognition; by refining and data enhancement processing the acquired corn image data, the stability of the recognition result is further improved; at the same time, according to the boundary box information, the corn plants are tracked to set an independent label and mark each corn plant, and the local traits are identified by combining the local trait recognition model based on the neural network, which can accurately identify the local area of ​​the corn plant, such as identifying the number of aerial root layers layer by layer, thereby effectively improving the accuracy of identifying the target traits of the corn plant. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the invention and, together with the description, serve to explain the principles of the invention.

[0015] Figure 1 A flowchart of a method for obtaining local traits of corn provided by an embodiment of the present invention; Figure 2 A technical roadmap for a method for obtaining local traits of corn provided by an embodiment of the present invention; Figure 3 A schematic diagram of a computer device for a method for obtaining local traits of corn provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0016] To make the objectives, technical solutions, and advantages of the present invention more clear, the technical solutions of the present invention will be clearly and completely described below in conjunction with the accompanying drawings. Obviously, the embodiments described are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts shall fall within the scope of protection of the present invention.

[0017] The following describes in detail the technical solution of the present invention and how the technical solution of the present invention solves the above-mentioned technical problems using specific embodiments. The following specific embodiments can be combined with each other, and the same or similar concepts or processes may not be repeated in some embodiments. The following embodiments of the present invention are described in conjunction with the accompanying drawings.

[0018] Figure 1 This is a flow chart of a method for obtaining local traits of corn provided by an embodiment of the present invention. Figure 1 As shown, this embodiment shows a method for obtaining local traits of corn, including: S1: Obtain historical field corn plant images and construct a corn image dataset.

[0019] Optionally, a field robot is used to obtain historical field corn plant images.

[0020] For example, the brightness and contrast of the corn image dataset may be adjusted to achieve image enhancement processing on the corn image dataset.

[0021] S2: Input the corn image dataset into a recognition model based on a neural network, train the recognition model, and obtain a preliminary corn plant recognition model for identifying corn characteristics.

[0022] Exemplarily, the neural network may be a YOLOV10 network.

[0023] S3: Obtain corn images from multiple test areas to construct a fine-tuning dataset, fine-tune the preliminary corn plant recognition model, and obtain a plant bounding box recognition model.

[0024] Optionally, when fine-tuning the preliminary corn plant recognition model, set the learning rate to 0.0005 and freeze Backbone.

[0025] For example, during the training of the model, the model parameters can be continuously adjusted through the back-propagation algorithm to improve the recognition accuracy of the model.

[0026] For example, the preliminary corn plant recognition model obtained in the above steps can be fine-tuned based on the initial weight file. Specific parameters include: learning rate lr = 0.0005, frozen backbone (only the detection head is trained). Images are enhanced and labeled using synchronously acquired sensor parameters. A relatively accurate corn recognition model weight file was trained using nearly 1,000 representative labeled photos. High-resolution images or video data of corn fields are collected using drones or MaizrobRobot field robots equipped with cameras. By combining synchronously acquired sensor parameters (such as light intensity and humidity) with the image or video data, the image enhancement process is optimized, thereby improving the distinctiveness of the training data. This allows the system to more accurately address quality issues in complex environments and further enhances its adaptability to dynamic scenarios.

[0027] S4: The corn images to be identified on both sides of the test area collected in real time by the field robot are input into the plant border recognition model to obtain the bounding box information of each corn plant in each frame of the corn image.

[0028] For example, the plant border recognition model can be deployed on a field robot equipped with a camera, and then the field robot equipped with the plant border recognition model is used to obtain corn images in the area to be tested.

[0029] S5: Based on the bounding box information of each corn plant, track the trajectory of each corn plant and set an independent mark, identify the target part image of the marked corn plant, and determine the corn local trait dataset to be identified.

[0030] For example, the target parts in the image can be identified and accurately separated by selecting the YOLO algorithm. For example, the root area of ​​the marked corn plant can be identified by the YOLOv10 algorithm, which serves as the basis for identifying root traits in subsequent steps, such as the number of aerial root layers.

[0031] S6: Construct a local trait recognition model based on a neural network; input the corn local trait dataset to be identified into the trained local trait recognition model, determine multiple local trait recognition results of the corn plant with the same mark in each frame, and select the local trait recognition result with the highest confidence as the final corn local trait recognition result.

[0032] For example, the key targets in the corn image can be obtained by identifying the image to be inspected, and the detected key targets can be tracked by using the BoT-SORT algorithm and the ByteTrack algorithm, and a unique corn label can be added to each corn plant. At the same time, the area detection algorithm of the double-line collision method can be used to perform real-time trajectory tracking and count the number of each corn plant on both sides of the area to be inspected where the Maizrob Robot is located.

[0033] Taking the identification of aerial root layers as an example, the YOLOv10 algorithm was first used to perform preliminary processing on the plant image, accurately segmenting the root region. Then, within the extracted root image, the specific number of aerial root layers was identified layer by layer. This multi-level segmentation and refinement process significantly improved the model's accuracy and robustness in trait identification tasks.

[0034] Exemplarily, the area detection algorithm of the double-line collision method may include setting the following two detection lines: line1=[(10,650),(1910,650)], line2=[(400,500),(1520,500)], and then counting the corn plants whose trajectories collide with the lines to obtain the number of corn plants in the field where the field robot is located.

[0035] For example, the YOLOv10 algorithm can be used to detect the root photos of each corn plant. If the corn plant does not appear in 10 consecutive frames, it is determined to have disappeared from the field of view. Each root photo of the corn plant is identified using a pre-trained root weight file to obtain the confidence level of 0, 1, 2, or 3 layers of aerial roots. The confidence levels of all root photos of the same corn plant within the field of view are summed up, and the plant with the highest overall confidence level is determined to have the highest number of aerial root layers. For example, if the confidence levels are summed up and the overall confidence level of 2 layers is the highest, then the corn plant has 2 layers of aerial roots. This allows us to determine the total number of corn plants in the field where the field robot is located, the number of aerial root layers of each corn plant, the proportion of plants with 0, 1, 2, or 3 layers of aerial roots, and the distribution of the number of aerial roots over the field robot's operating time.

[0036] For example, the segmentation results of a corn image may include, for example, the number of aerial root layers, plant height, and leaf area. When obtaining other traits in a corn image, the detection can be performed through the same steps as the aerial root layer detection model; specifically, the root photo capture position can be changed: according to the part of the corn where the trait is located, the photo capture position can be changed, such as the upper 1 / 4 of the plant for corn fruit-related traits. To train the corresponding local trait recognition model, it is only necessary to recognize the target part image. If YOLOv10 is used, a training set consisting of 400 labeled photos can produce good recognition results. Compared with the traditional method of direct recognition in the video, this greatly reduces the training cost and improves the recognition accuracy.

[0037] For example, a video or live broadcast source can be selected on the constructed visual local platform, the corn image to be identified can be input into the optimal corn plant recognition model, and the combined BoT-SORT algorithm or ByteTrack algorithm can be selected to identify, track, output bounding box coordinates and independently mark the corn plants; wherein, the visual local platform can include a password verification interface, and three recognition methods can be selected: picture, video, and live broadcast source; the platform can choose the minimum confidence level displayed, whether to enable tracking, enable BoT-SORT algorithm or ByteTrack algorithm to implement tracking and other model functions; the platform can choose whether to retain frame images with identification boxes and trajectories.

[0038] The above is a method for obtaining local traits of corn provided in one or more embodiments of this specification. Based on the same idea, this specification also provides a corresponding system for obtaining local traits of corn, including: The acquisition module is used to obtain historical field corn plant images and construct a corn image dataset; A corn plant preliminary recognition model determination module is used to input the corn image data set into a recognition model based on a neural network, train the recognition model, and obtain a corn plant preliminary recognition model for identifying corn characteristics; The plant border recognition model determination module is used to obtain corn images of multiple test areas to construct a fine-tuning dataset, fine-tune the preliminary corn plant recognition model, and obtain a plant border recognition model; The bounding box information determination module is used to input the corn images to be identified on both sides of the test area collected in real time by the field robot into the plant border recognition model to obtain the bounding box information of each corn plant in each frame of corn image; The marking module is used to track the trajectory of each corn plant and set an independent mark based on the bounding box information of each corn plant, identify the target part image of the marked corn plant, and determine the corn local trait dataset to be identified; The recognition module is used to build a local trait recognition model based on a neural network; the corn local trait dataset to be identified is input into the trained local trait recognition model, multiple local trait recognition results of the corn plant with the same mark in each frame are determined, and the local trait recognition result with the highest confidence is selected as the final corn local trait recognition result.

[0039] The specific definition of the corn local trait acquisition system can be found in the definition of the corn local trait acquisition method above and will not be repeated here. Each module in the above-mentioned corn local trait acquisition system can be implemented in whole or in part through software, hardware, or a combination thereof. Each of the above-mentioned modules can be embedded in or independent of the processor of the computer device in hardware form, or can be stored in the memory of the computer device in software form, so that the processor can call and execute the corresponding operations of each of the above modules.

[0040] The present invention also provides a computer-readable storage medium, which stores a computer program. The computer program can be used to execute the above-mentioned method for obtaining local traits of corn.

[0041] The present invention also provides Figure 3 The structural diagram of the computer equipment shown in FIG. Figure 3 As shown, at the hardware level, the computer device includes a processor, an internal bus, a network interface, memory, and non-volatile storage, and may also include other hardware required for operations. The processor reads the corresponding computer program from the non-volatile storage into the internal memory and then runs it to implement the method for obtaining local traits of corn provided in the above embodiment.

[0042] The technical features of the above embodiments can be combined arbitrarily. In order to make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of the present invention.

Claims

1. A method for obtaining local traits of corn, characterized in that: include: Obtain historical field corn plant images and construct a corn image dataset; Inputting the corn image dataset into a neural network-based recognition model, training the recognition model, and obtaining a preliminary corn plant recognition model for identifying corn characteristics; Obtain corn images from multiple test areas to construct a fine-tuning dataset, fine-tune the preliminary corn plant recognition model, and obtain a plant bounding box recognition model; The corn images to be identified on both sides of the test area collected in real time by the field robot are input into the plant border recognition model to obtain the bounding box information of each corn plant in each frame of corn image; Based on the bounding box information of each corn plant, each corn plant is tracked and independently marked, and the target part image of the marked corn plant is identified to determine the corn local trait dataset to be identified; A local trait recognition model based on a neural network was constructed. The corn local trait dataset to be identified was input into the trained local trait recognition model to determine multiple local trait recognition results of the corn plant with the same mark in each frame, and the local trait recognition result with the highest confidence was selected as the final corn local trait recognition result.

2. The method for obtaining local traits of corn according to claim 1, characterized in that: Tracking each corn plant and setting an independent marker based on the bounding box information of each corn plant specifically includes: The bounding box information of each corn plant is input into the BoT-SORT algorithm and ByteTrack algorithm to track the trajectory of the corn plant and obtain the corresponding trajectory information; According to the trajectory information, the double-line collision method is used to determine whether the area to which the corn plant belongs is the test area where the field robot is located. If so, a unique label is assigned to each corn plant in the current area.

3. The method for obtaining local traits of corn according to claim 1, wherein: The YOLOv10 algorithm is used to identify the target part image of the marked corn plant.

4. The method for obtaining local traits of corn according to claim 1, wherein: The neural network is the YOLOV10 network.

5. The method for obtaining local traits of corn according to claim 1, wherein: The acquisition of historical field corn plant images and construction of a corn image dataset specifically includes: The historical field corn plant images were extracted at intervals of 10 frames, and 3-4 corn plants on both sides of the ridge were manually marked to construct a corn image dataset.

6. The method for obtaining local traits of corn according to claim 1, wherein: An image enhancement operation is performed on the corn image dataset.

7. A system for obtaining local traits of corn, characterized in that: include: The acquisition module is used to obtain historical field corn plant images and construct a corn image dataset; A corn plant preliminary recognition model determination module is used to input the corn image data set into a recognition model based on a neural network, train the recognition model, and obtain a corn plant preliminary recognition model for identifying corn characteristics; The plant border recognition model determination module is used to obtain corn images of multiple test areas to construct a fine-tuning dataset, fine-tune the preliminary corn plant recognition model, and obtain a plant border recognition model; The bounding box information determination module is used to input the corn images to be identified on both sides of the test area collected in real time by the field robot into the plant border recognition model to obtain the bounding box information of each corn plant in each frame of corn image; The marking module is used to track the trajectory of each corn plant and set an independent mark based on the bounding box information of each corn plant, identify the target part image of the marked corn plant, and determine the corn local trait dataset to be identified; The recognition module is used to build a local trait recognition model based on a neural network; the corn local trait dataset to be identified is input into the trained local trait recognition model, multiple local trait recognition results of the corn plant with the same mark in each frame are determined, and the local trait recognition result with the highest confidence is selected as the final corn local trait recognition result.

8. A computer-readable storage medium, characterized in that The storage medium stores a computer program, and when the computer program is executed by a processor, the method for obtaining local traits of corn according to any one of claims 1 to 6 is implemented.

9. A computer device, characterized in that: The method comprises a memory, a processor and a computer program stored in the memory and capable of running on the processor, wherein when the processor executes the program, the method for obtaining local traits of corn according to any one of claims 1 to 6 is implemented.

Images