Corn seedling identification method and system based on remote sensing image and related equipment

Abstract

The invention relates to the technical field of image recognition, and discloses a corn seedling recognition method and system based on a remote sensing image and related equipment, and the method comprises the steps: obtaining the remote sensing image; performing scene filling on the remote sensing image not containing the preset scene by using the remote sensing image containing the preset scene, and determining an initial sample data set; labeling a preset scene of each remote sensing image in the initial sample data set to obtain a target sample data set; introducing the target sample data set into a target detection model for training, and determining a corn seedling identification model; and identifying the corn seedlings through the corn seedling identification model. According to the method, the remote sensing data not containing the preset scene is filled through the remote sensing image containing the preset scene, so that the number of the remote sensing images containing the preset scene and the number of the remote sensing images not containing the preset scene are balanced, the sample imbalance phenomenon is relieved, the data set quality is improved, and the corn seedling recognition precision is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of image recognition, and in particular to a method, system and related equipment for identifying corn seedlings based on remote sensing images. Background Art

[0002] As a vital food crop, corn cultivation and management are directly related to national food security. Accurately identifying corn seedlings provides essential data for efficient farmland management, enabling precise management of corn cultivation and improving production efficiency.

[0003] Existing corn seedling recognition mainly focuses on corn seedling recognition based on the YOLO target detection framework. For example, the patent published by Jiang Tiantian et al., "A method for calculating the number of corn seedlings based on YOLOv8", can accurately calculate the number of corn seedlings in complex field environments based on images of different resolutions. The patent published by Yin Xiaolin et al., "Corn seedling recognition method based on residual selective kernel attention mechanism", is based on the YOLOv8n algorithm and introduces a residual selective kernel attention mechanism to reduce the false detection rate and improve the accuracy of corn seedling detection. The patent published by Ma Huimin et al., "Corn field seedling counting method and system based on YOLOv8", captures global and local features in the image by adding a two-layer routing attention mechanism (BiFormer) to the YOLOv8 network, and adds spatial depth conversion convolution (SPD-Conv) to convert image spatial information into depth information, thereby improving the network's ability to learn image features. Boyi Tang et al. published a paper titled "Using UAV-Based Multispectral Imagery and the CGS-YOLO Algorithm to Distinguish Corn Seedlings from Weeds." This method constructs a corn seedling recognition algorithm (CGS-YOLO) for scenes with high weed coverage. This algorithm, also based on YOLO, incorporates a CARAFE sampling operation and a small object detection layer to extract contextual information between pixels, preserving the weak semantic features of corn seedling images.

[0004] Although the above method has improved the accuracy of corn seedling detection to a certain extent, the above technical method innovation focuses on the improvement of the algorithm model and ignores the improvement of the dataset preparation method, resulting in poor dataset quality and low accuracy of the algorithm model in identifying corn seedlings. Summary of the Invention

[0005] In order to overcome the problem that when using traditional machine learning methods to detect corn seedlings, improvements in dataset preparation methods are ignored, resulting in poor dataset quality and low accuracy of algorithm models for corn seedling recognition, the present invention provides a corn seedling recognition method, system and related equipment based on remote sensing images.

[0006] In a first aspect, in order to solve the above technical problems, the present invention provides a method for identifying corn seedlings based on remote sensing images, comprising:

[0007] Acquiring remote sensing images; wherein the remote sensing images are remote sensing images of corn seedlings collected in different field scenes;

[0008] Using remote sensing images containing preset scenes to fill in scenes in remote sensing images that do not contain preset scenes, an initial sample dataset is determined; wherein the preset scenes are complex scenes that affect corn seedling recognition;

[0009] Label the preset scene of each remote sensing image in the initial sample data set to obtain the target sample data set;

[0010] The target sample dataset is introduced into the target detection model for training to determine the corn seedling recognition model;

[0011] Corn seedlings are identified using a corn seedling identification model.

[0012] In a second aspect, the present invention provides a corn seedling identification system based on remote sensing images, comprising:

[0013] A remote sensing image acquisition module is used to acquire remote sensing images; wherein the remote sensing images are remote sensing images of corn seedlings collected in different field scenes;

[0014] An initial sample data set determination module is used to use remote sensing images containing preset scenes to fill in remote sensing images that do not contain preset scenes to determine an initial sample data set; wherein the preset scenes are complex scenes that affect corn seedling recognition;

[0015] The target sample data set determination module is used to annotate the preset scene of each remote sensing image in the initial sample data set to obtain the target sample data set;

[0016] A corn seedling recognition model determination module is used to introduce the target sample data set into the target detection model for training and determine the corn seedling recognition model;

[0017] The corn seedling recognition module is used to identify corn seedlings using a corn seedling recognition model.

[0018] In a third aspect, the present invention provides a computing device comprising a memory, a processor, and a program stored in the memory and running on the processor, wherein when the processor executes the program, the steps of the above-mentioned method for identifying corn seedlings based on remote sensing images are implemented.

[0019] In a fourth aspect, the present invention provides a computer-readable storage medium, which stores instructions. When the instructions are executed on a terminal device, the terminal device executes the steps of the above-mentioned corn seedling identification method based on remote sensing images.

[0020] The beneficial effects of the present invention are as follows: remote sensing images are acquired, and then remote sensing images that do not contain preset scenes are scene-filled with remote sensing images that contain preset scenes, thereby obtaining an initial sample data set, and then after labeling the initial sample data set, a training data set (target sample data set) for a target detection model is obtained, and finally corn seedlings are identified using a trained corn seedling recognition model. The present application uses remote sensing images that contain preset scenes to fill remote sensing data that do not contain preset scenes, thereby balancing the number of remote sensing images that contain preset scenes and the number of remote sensing images that do not contain preset scenes, alleviating sample imbalance, improving data set quality, and thereby improving corn seedling recognition accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the present invention is further described below with reference to the accompanying drawings and embodiments.

[0022] Figure 1 Schematic diagram of a process for identifying corn seedlings based on remote sensing images according to an embodiment of the present invention;

[0023] Figure 2 This is the recognition result of a film-mulching scene with more weeds;

[0024] Figure 3 This is the recognition result of a scene with a lot of weeds and severe lighting and shadows;

[0025] Figure 4 Recognition results for the wheat stubble scene;

[0026] Figure 5 Schematic diagram of the structure of a corn seedling identification system based on remote sensing images according to an embodiment of the present invention. DETAILED DESCRIPTION

[0027] The following examples are provided to further explain and supplement the present invention and do not constitute any limitation to the present invention.

[0028] The following describes a method, system, and related equipment for identifying corn seedlings based on remote sensing images according to an embodiment of the present invention with reference to the accompanying drawings.

[0029] like Figure 1 As shown, an embodiment of the present invention provides a method for identifying corn seedlings based on remote sensing images, comprising:

[0030] S1. Acquire remote sensing images; wherein the remote sensing images are remote sensing images of corn seedlings collected in different field scenes.

[0031] In this embodiment, a DJI Mavic 3E drone equipped with a 20-megapixel image sensor (RGB camera) is used to collect RGB images of 2-4 leaf corn seedlings in different field scenes, wherein the different field scenes include: spring corn mulching scene, weed coverage scene, bare soil scene, summer corn stubble high scene, weed coverage scene, large and small seedlings scene, and scenes with different light intensities. For scenes with more seedlings, scenes with higher stubble, and scenes with higher weed coverage, the drone flies at an altitude of 12 meters and an image resolution of 0.3 cm; for other scenes, the drone flies at an altitude of 26 meters and an image resolution of 0.7 cm. Using DJI Zhitu software, the captured images are reconstructed in two dimensions to obtain orthophotos in the geotiff format under the WGS1984 geographic coordinates, i.e., remote sensing images.

[0032] S2. Use remote sensing images containing preset scenes to fill in remote sensing images that do not contain preset scenes to determine an initial sample data set; wherein the preset scenes are complex scenes that affect corn seedling recognition.

[0033] In this embodiment, the preset scenes may include large and small seedling scenes, weed-covered scenes, and light and shadow scenes. In such scenes, corn seedlings may be obscured by large seedlings, weeds, and light and shadow scenes, resulting in the inability to identify the corn seedlings. In addition, if the ratio of the area of ​​debris or light and shadow to the total area of ​​the remote sensing image exceeds a threshold, it indicates that the preset scene is included. The threshold is set according to actual conditions. For example, if the ratio of the area of ​​debris or light and shadow to the total area of ​​the remote sensing image exceeds 0.3, it indicates that the remote sensing image contains the preset scene.

[0034] S3. Label the preset scene of each remote sensing image in the initial sample data set to obtain the target sample data set.

[0035] In this embodiment, the labelimg annotation tool is used to annotate the preset scenes in the sample graphics. The annotation format is set to YOLO format, and the single category annotation mode is set during annotation.

[0036] S4. Introduce the target sample data set into the target detection model for training to determine the corn seedling recognition model.

[0037] S5. Identify corn seedlings using a corn seedling identification model.

[0038] In this embodiment, remote sensing images are obtained, and then the remote sensing images that do not contain the preset scenes are filled with remote sensing images that contain the preset scenes, thereby obtaining an initial sample data set. After the initial sample data set is annotated, a training data set (target sample data set) for the target detection model is obtained. Finally, the trained corn seedling recognition model is used to identify corn seedlings. This application uses remote sensing images that contain the preset scenes to fill the remote sensing data that do not contain the preset scenes, thereby balancing the number of remote sensing images that contain the preset scenes and the number of remote sensing images that do not contain the preset scenes, alleviating the sample imbalance phenomenon, improving the data set quality, and thus improving the corn seedling recognition accuracy.

[0039] Optionally, using a remote sensing image containing a preset scene to fill a remote sensing image that does not contain a preset scene to determine an initial sample data set includes:

[0040] Screening out a first image containing a preset scene and a second image not containing the preset scene from the remote sensing image;

[0041] Cropping a preset scene in the first image into a first image block of a preset size, and randomly filling the first image block into a random position in the randomly selected second image to determine a third image;

[0042] The third image is used as the initial sample data set.

[0043] In this embodiment, in normal plots, the number of remote sensing images containing preset scenes is small, while the number of remote sensing images not containing preset scenes is far higher than that of remote sensing images not containing preset scenes. Therefore, by cutting the preset scenes into the remote sensing images not containing preset scenes, the sample imbalance phenomenon is effectively alleviated, the quality of the data set is improved, and the accuracy of corn seedling recognition is thereby improved.

[0044] Optionally, cropping the preset scene in the first image into a first image block of a preset size includes:

[0045] Projecting the first image using GIS software to determine a projected image;

[0046] Draw a vector point at the location of the preset scene in the projected image, and obtain the projection coordinates of the vector point in the projected image; wherein the vector point is drawn at the center of the preset scene;

[0047] Calculate the pixel coordinates of the vector point in the first image according to the coordinates of the upper left corner of the projected image, the resolution, and the projection coordinates;

[0048] An image of a preset size is cropped from the first image using the pixel coordinates of the vector point as a center point as a first image block.

[0049] In this example, ArcGIS software is used to open the geotiff image, and the "Data Management Tool>Projections and Transformations>Project" tool is used to convert the WGS1984 geographic coordinates of the remote sensing image into UTM projection coordinates; then a new vector point file is created and the editing function is enabled to mark the vector points of the preset scene.

[0050] In this embodiment, the preset size is set according to actual conditions. Preferably, in this embodiment, the preset size may range from 100*100 to 400*400.

[0051] The first image is read into array format using the OpenCV open source library. The array of the complex scene is randomly sliced ​​and cropped according to a preset size, with the slice size ranging from 100*100 to 400*400. The array slices are randomly filled into random positions in the randomly selected second image.

[0052] Optionally, the pixel coordinates of the vector point in the first image are calculated according to the coordinates of the upper left corner of the projected image, the resolution, and the projection coordinates, using the following formula:

[0053] (x p ,y p )=[(x g ,y g )-(X,Y)] / r

[0054] Among them, (x p ,y p ) represents pixel coordinates, (x g ,y g ) represents the projection coordinates, (X, Y) represents the coordinates of the upper left corner of the projected image, and r represents the resolution.

[0055] In this embodiment, the open source Python library pyshp is used to read the vector point file to obtain the projection coordinates of each vector point; the open source Python library gdal is used to read the projected geotiff image (projection image), obtain the upper left corner coordinates and resolution of the projection image, and calculate the pixel coordinates of the vector point on the remote sensing image.

[0056] Optionally, the target detection model is a YOLOv11n target detection model that has been pre-trained on the COCO dataset, wherein the learning rate of the YOLOv11n target detection model is set to a first preset value, and the number of training rounds is set to a second preset value. After the YOLOv11n target detection model is trained with the learning rate of the first preset value and the number of training rounds of the second preset value, if the verification accuracy of the YOLOv11n target detection model does not improve for a continuous preset number of rounds, the YOLOv11n target detection model at this time is used as a corn seedling recognition model, and the preset number of rounds is less than or equal to the second preset value.

[0057] In this embodiment, the first preset value, the second preset value, and the preset number of rounds are all set according to actual conditions. Preferably, in this embodiment, the first preset value can be 10. -2 , the second preset value can be 500, and the preset number of rounds can be 10 rounds.

[0058] In this embodiment, the target sample dataset can be randomly divided into an untrained set and a validation set in a ratio of 8:2. The YOLOv11n target detection model is trained using the training set. By selecting the YOLOv11n target detection model that has been pre-trained on the COCO dataset, the training cost is controlled and the model convergence is accelerated. When the verification accuracy of the YOLOv11n target detection model does not improve for 10 consecutive rounds, the validation set is used to verify the accuracy of the trained YOLOv11n target detection model. The early stopping strategy is used to prevent model overfitting. If the accuracy of the trained YOLOv11n target detection model meets the requirements, the YOLOv11n target detection model at this time is used as the corn seedling recognition model.

[0059] In this embodiment, YOLOv11 is one of the most efficient target detection algorithms. It has made significant improvements in architecture and training methods based on the previous YOLO version, especially in feature extraction technology, which can capture complex details more accurately. This makes YOLOv11 superior to previous versions such as YOLOv8 in detecting scenes with finer object details. YOLOv11n is the version with the smallest number of parameters among all YOLOv11 models, with low training cost and fast inference speed. It is particularly suitable for tasks of single-category target recognition that take into account both accuracy and efficiency. Therefore, the high-efficiency and high-precision YOLOv11n target detection model is introduced into the corn seedling recognition task, which can accurately identify corn seedlings in a variety of scenarios. The method is highly applicable and can effectively improve the recognition speed and accuracy of field scenes.

[0060] Optionally, identifying the corn seedlings using a corn seedling identification model includes:

[0061] Acquire a remote sensing image to be identified, and read the remote sensing image to be identified using a sliding window with a preset window size and a preset step length to determine a plurality of second image blocks;

[0062] Inputting each second image block into a corn seedling recognition model to determine a bounding box in each second image block;

[0063] Obtaining a coordinate position of a bounding box on the remote sensing image to be identified based on a positional relationship of the second image block on the remote sensing image to be identified;

[0064] Get the confidence of each bounding box, and calculate the intersection-over-union ratio between each bounding box based on the coordinate position;

[0065] Based on the confidence level and intersection-over-union ratio, a bounding box that meets the preset conditions is selected as the target bounding box; the target bounding box is the location of the corn seedling.

[0066] In this embodiment, the preset window size and the preset step size are set according to actual conditions. Preferably, in this embodiment, the preset window size is 640*640, and the preset step size is 512. When the sliding window acquires the second image block, the lateral overlap between two adjacent second image blocks is set to 20%.

[0067] In this embodiment, the GDAL open source library is used to read the remote sensing image to be identified, and the image is converted into an array format. The array is read in blocks using a sliding window method, and the pixel coordinate positions of the array on the remote sensing image to be identified are recorded, that is, the position relationship of the second image block on the remote sensing image to be identified. Based on the above pixel coordinate positions, the coordinate position of the bounding box on the remote sensing image to be identified can be calculated, thereby restoring the bounding box to the remote sensing image to be identified.

[0068] Optionally, based on the confidence level and the intersection-over-union ratio, a bounding box that meets preset conditions is selected as the target bounding box, including:

[0069] Filtering out a bounding box whose confidence level is higher than a third preset value and whose intersection-over-union ratio is lower than a preset threshold as a first target bounding box;

[0070] Filtering out a bounding box whose confidence level is higher than a third preset value and whose intersection-over-union ratio is higher than or equal to a preset threshold as a second target bounding box;

[0071] Deduplicating overlapping portions of any two overlapping second target bounding boxes to determine a third target bounding box;

[0072] The target bounding box includes a first target bounding box and a third target bounding box.

[0073] In this embodiment, the third preset value and the preset threshold are set according to actual conditions. Preferably, in this embodiment, the third preset value may be 0.3, and the preset threshold may be 0.1.

[0074] In this embodiment, after screening out the second target bounding boxes whose confidence is higher than the third preset value and whose intersection-over-union ratio is higher than or equal to the preset threshold, the nms method in the torchvision open source library is used to implement NMS deduplication, thereby deduplicating the overlapping parts between the two overlapping second target bounding boxes.

[0075] In this embodiment, the open source library PIL library is used to draw images corresponding to the first target bounding box and the second target bounding box on the remote sensing image to be recognized to obtain the final recognition result.

[0076] like Figure 2-Figure 4 As shown in the figure, the recognition results of spring corn and summer corn seedlings in a certain area are obtained in this embodiment. The resolution of the drone image is 0.7 cm. Figure 2 It can be seen that in the film-covered scene with more weeds, the corn seedlings are blocked by the weeds, but the corn seedlings are still accurately identified ( Figure 2 The middle arrow points to the enlarged image in the large image. From the enlarged image, you can see several marked target bounding boxes, such as m0.70, m0.66, etc.).

[0077] from Figure 3 It can be seen that in the scene with many weeds and severe light shadows, the corn seedlings are blocked by weeds and light noise shadows, but the corn seedlings are still accurately identified ( Figure 3 The middle arrow points to the enlarged image in the large image. From the enlarged image, you can see several marked target bounding boxes, such as m0.40, m0.47, etc.).

[0078] from Figure 4 It can be seen that in the wheat stubble scene, the corn seedlings are blocked by the wheat stubble, but the corn seedlings are still accurately identified ( Figure 4 The middle arrow points to the enlarged image in the large image. From the enlarged image, you can see several marked target bounding boxes, such as m0.61, m0.67, etc.).

[0079] like Figure 5 As shown, the present invention provides a corn seedling identification system based on remote sensing images, comprising:

[0080] A remote sensing image acquisition module is used to acquire remote sensing images; wherein the remote sensing images are remote sensing images of corn seedlings collected in different field scenes;

[0081] An initial sample data set determination module is used to use remote sensing images containing preset scenes to fill in remote sensing images that do not contain preset scenes to determine an initial sample data set; wherein the preset scenes are complex scenes that affect corn seedling recognition;

[0082] The target sample data set determination module is used to annotate the preset scene of each remote sensing image in the initial sample data set to obtain the target sample data set;

[0083] A corn seedling recognition model determination module is used to introduce the target sample data set into the target detection model for training and determine the corn seedling recognition model;

[0084] The corn seedling recognition module is used to identify corn seedlings using a corn seedling recognition model.

[0085] Optionally, the initial sample data set determination module is specifically configured to:

[0086] Screening out a first image containing a preset scene and a second image not containing the preset scene from the remote sensing image;

[0087] Cropping a preset scene in the first image into a first image block of a preset size, and randomly filling the first image block into a random position in the randomly selected second image to determine a third image;

[0088] The third image is used as the initial sample data set.

[0089] Optionally, the target sample data set determination module is specifically configured to:

[0090] Projecting the first image using GIS software to determine a projected image;

[0091] Draw a vector point at the location of the preset scene in the projected image, and obtain the projection coordinates of the vector point in the projected image; wherein the vector point is drawn at the center of the preset scene;

[0092] Calculate the pixel coordinates of the vector point in the first image according to the coordinates of the upper left corner of the projected image, the resolution, and the projection coordinates;

[0093] An image of a preset size is cropped from the first image using the pixel coordinates of the vector point as a center point as a first image block.

[0094] Optionally, the target sample data set determination module is specifically configured to:

[0095] According to the coordinates of the upper left corner of the projected image, the resolution and the projection coordinates, the pixel coordinates of the vector point in the first image are calculated using the following formula:

[0096] (x p ,y p )=[(x g ,y g )-(X,Y)] / r

[0097] Among them, (x p ,y p) represents pixel coordinates, (x g ,y g ) represents the projection coordinates, (X, Y) represents the coordinates of the upper left corner of the projected image, and r represents the resolution.

[0098] Optionally, the corn seedling identification module is specifically used to:

[0099] Acquire a remote sensing image to be identified, and read the remote sensing image to be identified using a sliding window with a preset window size and a preset step length to determine a plurality of second image blocks;

[0100] Inputting each second image block into a corn seedling recognition model to determine a bounding box in each second image block;

[0101] Obtaining a coordinate position of a bounding box on the remote sensing image to be identified based on a positional relationship of the second image block on the remote sensing image to be identified;

[0102] Get the confidence of each bounding box, and calculate the intersection-over-union ratio between each bounding box based on the coordinate position;

[0103] Based on the confidence level and intersection-over-union ratio, a bounding box that meets the preset conditions is selected as the target bounding box; the target bounding box is the location of the corn seedling.

[0104] Optionally, the corn seedling identification module is specifically used to:

[0105] Filtering out a bounding box whose confidence level is higher than a third preset value and whose intersection-over-union ratio is lower than a preset threshold as a first target bounding box;

[0106] Filtering out a bounding box whose confidence level is higher than a third preset value and whose intersection-over-union ratio is higher than or equal to a preset threshold as a second target bounding box;

[0107] Deduplicating overlapping portions of any two overlapping second target bounding boxes to determine a third target bounding box;

[0108] The target bounding box includes a first target bounding box and a third target bounding box.

[0109] An embodiment of the present invention also provides a computing device, including a memory, a manager, and a program stored in the memory and running on the manager. When the manager executes the program, it implements some or all steps of the above-mentioned corn seedling identification method based on remote sensing images.

[0110] Among them, the computing device can be a computer, and correspondingly, its program is computer software. The above-mentioned parameters and steps in the computing device of the present invention can refer to the parameters and steps in the embodiment of the corn seedling identification method based on remote sensing images above, and will not be repeated here.

[0111] Those skilled in the art will appreciate that the present invention may be implemented as a system, method, or computer program product. Therefore, the present disclosure may be specifically implemented in the following forms, namely: in the form of complete hardware, complete software (including firmware, resident software, microcode, etc.), or a combination of hardware and software, generally referred to herein as a "circuit," "module," or "system." Furthermore, in some embodiments, the present invention may also be implemented in the form of a computer program product in one or more computer-readable media, the computer-readable media containing computer-readable program code. Computer-readable storage media may be, for example, but not limited to, electrical, magnetic, optical, electromagnetic, infrared, or semiconductor systems, devices, or components, or any combination thereof.

[0112] In the description of this specification, the reference terms "one embodiment", "some embodiments", "example", "specific example", or "some examples" mean that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described can be combined in any one or more embodiments or examples in a suitable manner. In addition, those skilled in the art can combine and combine different embodiments or examples described in this specification and features of different embodiments or examples without contradiction.

[0113] Although the embodiments of the present invention have been shown and described above, it will be understood that the above embodiments are illustrative and are not to be construed as limitations on the present invention. A person skilled in the art may change, modify, replace and modify the above embodiments within the scope of the present invention.

Claims

1. A corn seedling recognition method based on remote sensing images, characterized in that: include: Acquiring remote sensing images; wherein the remote sensing images are remote sensing images of corn seedlings collected in different field scenes; Using a remote sensing image containing a preset scene to fill a remote sensing image that does not contain the preset scene to determine an initial sample data set; wherein the preset scene is a complex scene that affects the recognition of corn seedlings; Labeling the corn seedlings in the initial sample data set to obtain a target sample data set; The target sample data set is introduced into the target detection model for training to determine a corn seedling recognition model; Corn seedlings are identified using the corn seedling identification model.

2. The method according to claim 1, characterized in that The method of using the remote sensing image containing the preset scene to fill the remote sensing image that does not contain the preset scene to determine the initial sample data set includes: screening out a first image containing the preset scene and a second image not containing the preset scene from the remote sensing image; Cropping a preset scene in the first image into a first image block of a preset size, and randomly filling the first image block into a randomly selected random position in the second image to determine a third image; The third image is used as the initial sample data set.

3. The method according to claim 1, characterized in that The step of cropping the preset scene in the first image into a first image block of a preset size includes: Projecting the first image using GIS software to determine a projected image; Draw a vector point at the location of the preset scene in the projected image, and obtain the projection coordinates of the vector point in the projected image; wherein the vector point is drawn at the center of the preset scene; Calculating the pixel coordinates of the vector point in the first image according to the coordinates of the upper left corner of the projected image, the resolution, and the projection coordinates; An image of a preset size is cropped from the first image using the pixel coordinates of the vector point as a center point as a first image block.

4. The method according to claim 3, characterized in that The pixel coordinates of the vector point in the first image are calculated according to the coordinates of the upper left corner of the projected image, the resolution, and the projection coordinates, using the following formula: (x p ,and p )=[(x g ,and g )-(X,Y)] / r Among them, (x p ,y p ) represents pixel coordinates, (x g ,y g ) represents the projection coordinates, (X, Y) represents the coordinates of the upper left corner of the projected image, and r represents the resolution.

5. The method according to claim 1, characterized in that The target detection model is a YOLOv11n target detection model that has been pre-trained on the COCO dataset, wherein the learning rate of the YOLOv11n target detection model is set to a first preset value, and the number of training rounds is set to a second preset value. After the YOLOv11n target detection model is trained with the learning rate of the first preset value and the number of training rounds of the second preset value, if the verification accuracy of the YOLOv11n target detection model does not improve for a continuous preset number of rounds, the YOLOv11n target detection model at this time is used as a corn seedling recognition model, and the preset number of rounds is less than or equal to the second preset value.

6. The method according to claim 1, characterized in that The identifying of corn seedlings by using the corn seedling identification model includes: Acquire a remote sensing image to be identified, and read the remote sensing image to be identified using a sliding window with a preset window size and a preset step length to determine a plurality of second image blocks; inputting each of the second image blocks into the corn seedling recognition model to determine a bounding box in each of the second image blocks; Obtaining a coordinate position of the bounding box on the remote sensing image to be identified based on a positional relationship of the second image block on the remote sensing image to be identified; Obtaining the confidence level of each of the bounding boxes, and calculating the intersection-over-union ratio between any two of the bounding boxes based on the coordinate positions; According to the confidence level and the intersection-over-union ratio, a bounding box that meets preset conditions is selected as a target bounding box; wherein the target bounding box is the location of the corn seedling.

7. The method according to claim 6, characterized in that The step of selecting a bounding box that meets preset conditions as a target bounding box according to the confidence level and the intersection-over-union ratio includes: Filtering the bounding box whose confidence level is higher than a third preset value and whose intersection-over-union ratio is lower than a preset threshold as the first target bounding box; Filtering out the bounding box whose confidence level is higher than a third preset value and whose intersection-over-union ratio is higher than or equal to a preset threshold as the second target bounding box; Deduplicating overlapping portions of any two of the second target bounding boxes to determine a third target bounding box; The target bounding box includes the first target bounding box and the third target bounding box.

8. A corn seedling identification system based on remote sensing images, characterized in that: include: A remote sensing image acquisition module, configured to acquire remote sensing images; wherein the remote sensing images are remote sensing images of corn seedlings collected in different field scenes; An initial sample data set determination module is configured to use a remote sensing image containing a preset scene to perform scene filling on a remote sensing image that does not contain the preset scene, thereby determining an initial sample data set; wherein the preset scene is a complex scene that affects corn seedling recognition; a target sample data set determination module, configured to annotate the preset scene of each remote sensing image in the initial sample data set to obtain a target sample data set; A corn seedling recognition model determination module is used to introduce the target sample data set into the target detection model for training and determine the corn seedling recognition model; The corn seedling identification module is used to identify the corn seedlings using the corn seedling identification model.

9. A computing device comprising a memory, a processor, and a program stored in the memory and running on the processor, characterized in that: When the processor executes the program, the steps of the corn seedling identification method based on remote sensing images as described in any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium, characterized in that The computer-readable storage medium stores instructions, which, when executed on a terminal device, cause the terminal device to execute the steps of the method for identifying corn seedlings based on remote sensing images as described in any one of claims 1 to 7.

Images