A crop breeding operation method, device and medium based on a large model

Abstract

The application provides a crop breeding operation method and device based on a large model and a medium, and belongs to the technical field of crop breeding. The method obtains a first crop image and corresponding crop breeding information; wherein the crop breeding information at least includes a key part of the crop, a gene sequence, and a trait expression characteristic; a data set corresponding to the first crop image and the crop breeding information is generated and sent to a preset GPU cluster to train a selection large model deployed on the GPU cluster; a second crop image from a UAV executing a selection operation task is obtained; through a preset interface, the trained selection large model is called to identify the second crop image; in the case that the second crop image meets a corresponding selection target condition of the selection operation task, position information corresponding to the second crop image is sent to a user terminal.

Description

Technical Field

[0001] This application relates to the field of crop breeding technology, and in particular to a crop breeding operation method, equipment and medium based on a large model. Background Technology

[0002] With economic development and improved living standards, people have increasingly higher demands for crops that are closely related to their health, safety, and quality of life. These demands include crop yield, disease resistance, taste, and storage capacity.

[0003] To meet people's diverse needs for crops, large-scale crop breeding is imperative. Large-scale crop breeding differs from field crop cultivation; it requires breeders to frequently visit breeding bases to collect crop data on various crops and conduct corresponding research in laboratories.

[0004] Breeders need to inspect each crop at the breeding base and select those to be studied. This process is highly repetitive and complex, consuming a great deal of the breeders' research time and severely impacting the efficiency of breeding work. Summary of the Invention

[0005] To address the aforementioned issues, embodiments of this application provide a method, equipment, and medium for crop breeding operations based on a large model.

[0006] On the one hand, embodiments of this application provide a crop breeding operation method based on a large model, the method comprising:

[0007] Acquire a first crop image and corresponding crop breeding information; wherein the crop breeding information includes at least key parts of the crop, gene sequences, and trait expression characteristics;

[0008] Generate a dataset corresponding to the first crop image and the crop breeding information, and send it to a pre-configured GPU cluster to train a large-scale seed selection model deployed on the GPU cluster;

[0009] Acquire second crop images from a drone performing seed selection tasks;

[0010] The second crop image is identified by calling the trained seed selection model through a preset interface.

[0011] If the second crop image meets the seed selection target conditions corresponding to the seed selection task, the location information corresponding to the second crop image will be sent to the user terminal.

[0012] In one implementation of this application, generating a dataset corresponding to the first crop image and the crop breeding information specifically includes:

[0013] The first crop image is preprocessed; the preprocessing includes noise reduction, contrast enhancement, and image segmentation.

[0014] A set of crop features in the preprocessed first crop image is determined by a preset feature extraction algorithm; the set of crop features includes at least the following trait expression features: crop shape, crop texture, and crop color;

[0015] The dataset is generated based on the key parts of the crop, the set of crop features, and the corresponding gene sequences in the first crop image pre-annotated by the user.

[0016] In one implementation of this application, the seed selection task originates from the user terminal wirelessly connected to the drone; the seed selection task includes a seed selection area and a seed selection path.

[0017] In one implementation of this application, the trained seed selection model is invoked through a preset interface to recognize the second crop image, specifically including:

[0018] The seed selection model is used to determine the identification and breeding information corresponding to the second crop image; the seed selection model is a multimodal pre-trained model Unicoder-VL.

[0019] The identified breeding information is matched with the target breeding information corresponding to the selection target;

[0020] If the similarity between the identified breeding information and the target breeding information is greater than a preset threshold, the corresponding second crop image is determined to meet the selection target conditions;

[0021] Otherwise, the selection criteria are not met.

[0022] In one implementation of this application, the method further includes:

[0023] Once it is determined that the corresponding second crop image meets the seed selection target conditions, the acquisition location coordinates corresponding to the second crop image are determined; wherein, the acquisition location coordinates are sent by the UAV in real time;

[0024] Based on the collected location coordinates, a corresponding transformation relationship tuple is determined; wherein, the transformation relationship tuple includes coefficients used to convert the coordinates of each pixel point of the second crop image into the coordinates of the corresponding cultivation base location;

[0025] Determine the calibration pixel boxes in the second crop image that satisfy the selection target conditions; the calibration pixel boxes include one or more crops that satisfy the selection target conditions;

[0026] Based on the coordinates of each pixel corresponding to the calibrated pixel box and the transformation relationship tuple, the corresponding set of location coordinates of the cultivation base is determined, which is the location information.

[0027] In one implementation of this application, the method further includes:

[0028] When there are multiple seed selection tasks, the second crop image is received according to the corresponding task execution sequence;

[0029] If the interval between the reception of two second crop images is longer than a predetermined time, and the acquisition location coordinates of the two second crop images are respectively located in the cultivation base areas of two seed selection targets, it is determined that the drone has performed a transfer behavior to perform the seed selection operation task; the transfer behavior is that the drone performs different seed selection operation tasks.

[0030] The second crop image before the transfer action is added to the second crop image set corresponding to the first seed selection task, and the second crop image after the transfer action is added to the second crop image set corresponding to the second seed selection task, until the second crop image set for each seed selection task is obtained.

[0031] In one implementation of this application, the method further includes:

[0032] If none of the second crop images in the second crop image set meet the seed selection target conditions, an area expansion prompt is generated; the area expansion prompt is used to prompt the user to update the seed selection operation area corresponding to the seed selection operation task.

[0033] If the user sends feedback information indicating an area update based on the area expansion prompt, the image from the UAV in the updated seed selection operation area is used as the second crop image to determine whether the second crop image meets the seed selection target conditions; wherein, the feedback information includes at least the updated seed selection operation area.

[0034] In one implementation of this application, the method further includes:

[0035] When multiple drones perform the same seed selection task, a seed selection sub-path is determined for each drone in the seed selection area; the seed selection sub-paths are then spliced ​​together to obtain the seed selection path corresponding to the seed selection task; the seed selection sub-path includes path nodes for capturing images of the second crop.

[0036] The path node that is closest to the seed selection operation sub-path in each of the seed selection operation sub-paths is used as the splicing node; a seed selection operation sub-path includes one or two splicing nodes;

[0037] The second crop images acquired by each of the stitching nodes are stitched together according to the seed selection operation path, and the corresponding stitched images are used as the second crop images.

[0038] On the other hand, embodiments of this application also provide a crop breeding operation device based on a large model, the device comprising:

[0039] At least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to:

[0040] Acquire a first crop image and corresponding crop breeding information; wherein the crop breeding information includes at least key parts of the crop, gene sequences, and trait expression characteristics;

[0041] Generate a dataset corresponding to the first crop image and the crop breeding information, and send it to a pre-configured GPU cluster to train a large-scale seed selection model deployed on the GPU cluster;

[0042] Acquire second crop images from a drone performing seed selection tasks;

[0043] The second crop image is identified by calling the trained seed selection model through a preset interface.

[0044] If the second crop image meets the seed selection target conditions corresponding to the seed selection task, the location information corresponding to the second crop image will be sent to the user terminal.

[0045] Furthermore, embodiments of this application also provide a non-volatile computer storage medium storing computer-executable instructions, wherein the computer-executable instructions are configured as follows:

[0046] Acquire a first crop image and corresponding crop breeding information; wherein the crop breeding information includes at least key parts of the crop, gene sequences, and trait expression characteristics;

[0047] Generate a dataset corresponding to the first crop image and the crop breeding information, and send it to a pre-configured GPU cluster to train a large-scale seed selection model deployed on the GPU cluster;

[0048] Acquire second crop images from a drone performing seed selection tasks;

[0049] The second crop image is identified by calling the trained seed selection model through a preset interface.

[0050] If the second crop image meets the seed selection target conditions corresponding to the seed selection task, the location information corresponding to the second crop image will be sent to the user terminal.

[0051] Through the above scheme, this application can utilize a large-scale seed selection model deployed on a GPU cluster to identify second crop images collected by drones, thereby obtaining the crop location information corresponding to the seed selection target. This allows breeders to remotely understand the breeding results and obtain the location information of crops that can be used for research. This intelligently and flexibly reduces repetitive work for breeders and improves breeding efficiency. Attached Figure Description

[0052] The accompanying drawings, which are included to provide a further understanding of this application and form part of this application, illustrate exemplary embodiments and are used to explain this application, but do not constitute an undue limitation of this application. In the drawings:

[0053] Figure 1 This is a schematic flowchart of a crop breeding operation method based on a large model, as described in an embodiment of this application.

[0054] Figure 2 This is a schematic diagram of the system structure corresponding to a crop breeding operation method based on a large model in an embodiment of this application;

[0055] Figure 3 This is a schematic diagram of the structure of a crop breeding operation device based on a large model, as described in an embodiment of this application. Detailed Implementation

[0056] To make the objectives, technical solutions, and advantages of this application clearer, the technical solutions of this application will be clearly and completely described below in conjunction with specific embodiments and corresponding drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0057] Large-scale models refer to machine learning models with a large number of parameters and complex structures. These models can be applied to handle large-scale data and complex problems. Traditional machine learning models, such as logistic regression, decision trees, and Naive Bayes, are relatively small in scale and can only process small amounts of data. Deep learning models, on the other hand, can contain millions of parameters and process massive amounts of data. Ultra-large-scale deep learning models can even reach tens of billions of parameters, requiring supercomputers for training.

[0058] The advantages of large-scale models lie in their strong ability to handle large-scale data and complex problems, resulting in higher accuracy and performance. However, the common problems associated with large-scale model training, such as long training and inference times, large model size, high storage costs, and the need for greater computing power, can be effectively addressed through data and technology collaborations between large-scale breeding companies, research institutions, and industrial internet enterprises.

[0059] Currently, large-scale agricultural models are primarily applied to digital management in animal husbandry, intelligent agricultural manufacturing, and production processing. Large-scale crop breeding differs from field crop cultivation; it requires not only the establishment of datasets of images, genes, and trait expression, but also the use of high-resolution cameras to accurately identify plant differentiation and dominant traits, thereby improving the ability to breed superior varieties.

[0060] By establishing datasets of crop images, genes, and trait expressions through crop image and data acquisition, and developing large-scale crop selection models based on image-based big data modeling technology, we will develop online services and selection systems that combine drones and high-resolution cameras to screen dominant plants with specific varietal traits within a given operational area. This will promote the application of crop image data in breeding work, enabling more efficient and accurate crop trait analysis and improving the efficiency of high-quality seed breeding in large breeding enterprises and research institutions.

[0061] Based on this, this application provides a crop breeding operation method, equipment and medium based on a large model to solve the problems of repetitive, complex and heavy work and low efficiency in the current field of agricultural breeding through artificial selection.

[0062] The various embodiments of this application are described in detail below with reference to the accompanying drawings.

[0063] This application provides a crop breeding operation method based on a large model, such as... Figure 1 As shown, the method may include steps S101-S105:

[0064] S101, the server obtains the first crop image and the corresponding crop breeding information.

[0065] Crop breeding information includes at least key parts of the crop, gene sequences, and trait expression characteristics.

[0066] It should be noted that the server, as the executor of the crop breeding operation method based on a large model, is merely an example; the executor is not limited to the server, and this application does not specifically limit it. The server is used to receive, store, and analyze image data such as first crop images and second crop images, as well as to receive target breeding information corresponding to the selection task and identify the selection target. Furthermore, the server is wirelessly connected to the UAV acquisition terminal and wired or wirelessly connected to the user terminal; the user terminal is capable of running an operation task management system platform that sends selection tasks to the UAV and the server.

[0067] The first crop image can be an image of the crop's leaves, flowers, fruits, etc., captured using a high-resolution camera or scanner. During the acquisition process, it is necessary to ensure that the images are clear and the colors are accurate, and to record the attribute information of each image, such as variety and growing environment. Crop breeding information can be a table pre-generated by the user showing the correspondence between data such as key crop parts, gene sequences, and trait expression characteristics.

[0068] S102, the server generates a dataset containing the first crop image and crop breeding information, and sends it to the pre-configured GPU cluster to train the large-scale seed selection model deployed on the GPU cluster.

[0069] In this embodiment of the application, generating a dataset corresponding to a first crop image and crop breeding information specifically includes:

[0070] The first crop image undergoes image preprocessing. Preprocessing includes denoising, contrast enhancement, and image segmentation. A pre-defined feature extraction algorithm is used to determine the crop feature set in the preprocessed first crop image. The crop feature set includes at least the following trait expression features: crop shape, crop texture, and crop color. A dataset is generated based on the key crop parts, crop feature set, and corresponding gene sequences in the user-annotated first crop image.

[0071] In other words, this application can perform image preprocessing on the first crop image using image processing software, such as denoising, contrast enhancement, and segmentation, to highlight the crop's features and details. Subsequently, crop feature extraction is performed, extracting features related to crop traits, such as shape, texture, and color, from the processed first crop image. Simultaneously, the key parts of the crop annotated by the user are determined, along with the crop feature set and corresponding gene sequences, to generate a dataset for training a large-scale seed selection model.

[0072] On a GPU cluster, this application allows for fine-tuning the Unicoder-VL model using supervised learning. Crop image data, traits, and genetic information are used as labels for training and optimization. After training, the fine-tuned model is evaluated using a test set to compare its performance in crop image recognition and gene prediction. Based on the evaluation results, further optimizations can be made, such as adjusting model parameters, increasing network depth, and increasing data diversity. Ultimately, a dedicated large-scale model based on crop data, namely the seed selection model, can be formed. This large-scale model is deployed on a GPU computing cluster to ensure sufficiently high running and inference speeds. Pre-defined inference interfaces are exposed to allow remote invocation and integration by seed selection system applications.

[0073] S103, the server acquires a second crop image from the drone performing the seed selection task.

[0074] In this embodiment, the seed selection task originates from a user terminal wirelessly connected to the drone. The seed selection task includes a seed selection area and a seed selection path.

[0075] The seed selection operation area may be, for example, area A or area B pre-specified by the user. The seed selection operation path may be generated by the user or generated by a preset path algorithm to capture images of various locations in the seed selection operation area. This application does not make any specific limitations on this.

[0076] S104, the server calls the trained seed selection model through a preset interface to identify the second crop image.

[0077] In this embodiment of the application, a pre-trained seed selection model is invoked through a preset interface to identify the second crop image, specifically including:

[0078] The selection model, a multimodal pre-trained model called Unicoder-VL, is used to determine the breeding information corresponding to the second crop image. The identified breeding information is then matched with the target breeding information corresponding to the selection objective. If the similarity between the identified breeding information and the target breeding information is greater than a preset threshold, the corresponding second crop image is determined to meet the selection objective conditions. Otherwise, it does not meet the selection objective conditions.

[0079] In other words, the server sends the second crop image to the seed selection model through a preset interface, and obtains the identification breeding information output by the model. This identification breeding information includes at least crop shape, texture, and color features. Simultaneously, the target breeding information is the crop characteristics specified by the user for the seed selection target. The server can calculate the cosine similarity between the two images. If the cosine similarity is greater than a preset threshold set by the user, the second crop image corresponding to the identified breeding information is considered to meet the seed selection target conditions; otherwise, it is not.

[0080] S105, if the second crop image meets the seed selection target conditions corresponding to the seed selection task, the server sends the location information corresponding to the second crop image to the user terminal.

[0081] User terminals can be understood as devices such as computers and mobile phones used by breeders; this application does not make any specific limitations on this.

[0082] In this embodiment of the application, determining the location information corresponding to the second crop image includes:

[0083] Once it is determined that the corresponding second crop image meets the selection target conditions, the coordinates of the acquisition location corresponding to the second crop image are determined. These acquisition location coordinates are transmitted in real-time by the UAV. Based on the acquisition location coordinates, a corresponding transformation relation tuple is determined. This transformation relation tuple includes coefficients used to convert the coordinates of each pixel point in the second crop image to the coordinates of the corresponding cultivation base location. The calibration pixel boxes in the second crop image that meet the selection target conditions are determined. These calibration pixel boxes include one or more crops that meet the selection target conditions. Based on the coordinates of each pixel point corresponding to the calibration pixel box and the transformation relation tuple, the corresponding set of cultivation base location coordinates is determined, which constitutes the location information.

[0084] In other words, the second crop image corresponds to a unique acquisition location coordinate when the drone captured the image; simultaneously, each acquisition location coordinate can be a path node of the drone, and each acquisition location coordinate can correspond to a transformation relation tuple, which can convert the location in the image to the actual location of the cultivation base. For example, [a, b], where a is the first coefficient used to transform the horizontal coordinate of the pixel coordinate point, and b is the second coefficient used to transform the vertical coordinate of the pixel coordinate point. When processing the second crop image, the server can generate a calibration pixel box in the second crop image when the selection target is identified, and select one or more crops that meet the selection target conditions. At the same time, it generates a set of cultivation base location coordinates of the selected crops, obtaining the actual geographical location of the selection target, so that users can go to the corresponding location for breeding work.

[0085] This saves breeders research time and costs, reduces repetitive work, and improves breeding efficiency.

[0086] In one embodiment of this application, multiple seed selection tasks may be issued at once. This application includes:

[0087] When there are multiple seed selection tasks, second crop images are received according to the corresponding task execution sequence. If the interval between receiving two second crop images is longer than a predetermined time, and the coordinates of the drone's acquisition location corresponding to the two second crop images are located in the cultivation base areas of the two seed selection targets, it is determined that the drone has engaged in a transfer behavior to perform a seed selection task. The transfer behavior is the drone performing different seed selection tasks. The second crop images before the transfer behavior are added to the second crop image set corresponding to the first seed selection task, and the second crop images after the transfer behavior are added to the second crop image set corresponding to the second seed selection task, until the second crop image sets for each seed selection task are obtained.

[0088] In other words, the server can sequentially acquire the second crop images corresponding to each seed selection task, and identify different seed selection tasks when the transition behavior of the above two seed selection tasks is met. The predetermined duration can be set by the user. That is, after completing the shooting of one seed selection task, the drone will take the second crop image of the next seed selection task after the predetermined duration. At the same time, the server will further verify the acquisition location coordinates to ensure accurate identification of the transition behavior. Thus, a set of second crop images corresponding to each seed selection task is obtained.

[0089] In one embodiment of this application, the seed selection area of ​​the seed selection task published by the user may have errors, therefore this application also includes:

[0090] If none of the second crop images in the second crop image set meet the seed selection target conditions, a region expansion prompt is generated. This prompt alerts the user to update the seed selection operation area corresponding to the seed selection task. If the user updates the area based on the feedback from the region expansion prompt, the image from the UAV within the updated seed selection operation area is used as the second crop image to determine whether the second crop image meets the seed selection target conditions. The feedback information must at least include an updated seed selection operation area.

[0091] In other words, the server can send a notification message to the user terminal indicating an expansion of the seed selection area. The user can then re-specify the seed selection area through their terminal to generate feedback information indicating an updated area. If the area is not re-specified, the feedback message indicates an unupdated area, prompting the user to either execute the next seed selection task or return to the designated location. The feedback information is simultaneously sent to both the drone and the server, allowing the drone to acquire a second crop image within the updated seed selection area to identify the seed selection target.

[0092] Furthermore, a single seed selection task may be performed using multiple drones, including:

[0093] When multiple drones perform the same seed selection task, a seed selection sub-path is determined for each drone within the seed selection area. These sub-paths are then stitched together to form the final seed selection path for the task. Each sub-path includes path nodes for capturing images of the second crop. The path node closest to the sub-path to be stitched is selected as the stitching node. Each sub-path may contain one or two stitching nodes. The second crop images captured by each stitching node are then stitched together according to the seed selection path, and the resulting stitched image is used as the final second crop image.

[0094] In other words, users can specify multiple seed selection sub-paths for drones within the seed selection area, either manually or through path segmentation software. These sub-paths can then be combined to form the final seed selection path. The location where a drone captures the second crop image on a seed selection sub-path is designated as a path node. The server can obtain the adjacent seed selection sub-paths (A) of each drone and one or two other sub-paths used for stitching together to obtain the final seed selection path. It then identifies the path node closest to the adjacent sub-path along the seed selection path direction as the stitching node. The server acquires the second crop images captured by each stitching node and stitches the second crop images from adjacent stitching nodes according to the order of the seed selection path to obtain the final second crop image.

[0095] Image stitching can prevent situations where crop images are missed or a certain crop image is repeatedly identified when multiple drones are performing the same seed selection task, thus saving computing resources and improving crop recognition accuracy.

[0096] This application provides a schematic diagram of the operational system corresponding to the crop breeding operation method based on a large model, as shown in the figure below. Figure 2 As shown, the user terminal publishes a seed selection task through the UAV operation task management system. This includes inputting the seed selection target (201) to the seed selection data analysis system on the server, and sending the seed selection operation area and path (202) to the UAV acquisition terminal. The UAV acquisition terminal then sends a second crop image (203) to the server. Remote access to the large-scale seed selection model (204) is provided through seed selection data analysis.

[0097] Through the above scheme, this application can utilize a large-scale seed selection model deployed on a GPU cluster to identify second crop images collected by drones, thereby obtaining the crop location information corresponding to the seed selection target. This allows breeders to remotely understand the breeding results and obtain the location information of crops that can be used for research. This intelligently and flexibly reduces repetitive work for breeders and improves breeding efficiency.

[0098] Figure 3 A schematic diagram of a crop breeding operation device based on a large model is provided for an embodiment of this application, as shown below. Figure 3 As shown, the device includes:

[0099] At least one processor; and a memory communicatively connected to the at least one processor. The memory stores instructions executable by the at least one processor, which, when executed by the at least one processor, enable the at least one processor to:

[0100] Acquire a first crop image and corresponding crop breeding information. The crop breeding information includes at least key crop parts, gene sequences, and trait expression characteristics. Generate a dataset containing the first crop image and crop breeding information, and send it to a pre-configured GPU cluster to train a large-scale seed selection model deployed on the GPU cluster. Acquire a second crop image from the drone performing the seed selection task. Use a pre-defined interface to call the trained large-scale seed selection model to identify the second crop image. If the second crop image meets the seed selection target conditions corresponding to the seed selection task, send the location information corresponding to the second crop image to the user terminal.

[0101] This application embodiment also provides a non-volatile computer storage medium storing computer-executable instructions, wherein the computer-executable instructions are configured as follows:

[0102] Acquire a first crop image and corresponding crop breeding information. The crop breeding information includes at least key crop parts, gene sequences, and trait expression characteristics. Generate a dataset containing the first crop image and crop breeding information, and send it to a pre-configured GPU cluster to train a large-scale seed selection model deployed on the GPU cluster. Acquire a second crop image from the drone performing the seed selection task. Use a pre-defined interface to call the trained large-scale seed selection model to identify the second crop image. If the second crop image meets the seed selection target conditions corresponding to the seed selection task, send the location information corresponding to the second crop image to the user terminal.

[0103] The various embodiments in this application are described in a progressive manner. Similar or identical parts between embodiments can be referred to mutually. Each embodiment focuses on describing the differences from other embodiments. In particular, the device and medium embodiments are basically similar to the method embodiments, so the descriptions are relatively simple; relevant parts can be referred to the descriptions of the method embodiments.

[0104] The devices, media, and methods provided in this application are one-to-one correspondences. Therefore, the devices and media also have similar beneficial technical effects as their corresponding methods. Since the beneficial technical effects of the methods have been described in detail above, the beneficial technical effects of the devices and media will not be repeated here.

[0105] It should also be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0106] The above description is merely an embodiment of this application and is not intended to limit the scope of this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of the claims of this application.

Claims

1. A crop breeding operation method based on a large model, characterized in that, The method includes: Acquire a first crop image and corresponding crop breeding information; wherein the crop breeding information includes at least key parts of the crop, gene sequences, and trait expression characteristics; Generate a dataset corresponding to the first crop image and the crop breeding information, and send it to a pre-configured GPU cluster to train a large-scale seed selection model deployed on the GPU cluster; Acquire second crop images from a drone performing seed selection tasks; The second crop image is identified by calling the trained seed selection model through a preset interface. If the second crop image meets the seed selection target conditions corresponding to the seed selection task, the location information corresponding to the second crop image will be sent to the user terminal. The method further includes: When there are multiple seed selection tasks, the second crop image is received according to the corresponding task execution sequence; If the interval between the reception of two second crop images is longer than a predetermined time, and the acquisition location coordinates of the two second crop images are respectively located in the cultivation base areas of two seed selection targets, it is determined that the drone has performed a transfer behavior to perform the seed selection operation task; the transfer behavior is that the drone performs different seed selection operation tasks. The second crop image before the transfer action is added to the second crop image set corresponding to the first seed selection task, and the second crop image after the transfer action is added to the second crop image set corresponding to the second seed selection task, until the second crop image set for each seed selection task is obtained; If none of the second crop images in the second crop image set meet the seed selection target conditions, an area expansion prompt is generated; the area expansion prompt is used to prompt the user to update the seed selection operation area corresponding to the seed selection operation task. If the user sends feedback information indicating an area update based on the area expansion prompt, the image from the UAV in the updated seed selection operation area is used as the second crop image to determine whether the second crop image meets the seed selection target conditions; wherein, the feedback information includes at least the updated seed selection operation area.

2. The crop breeding operation method based on a large model according to claim 1, characterized in that, Generating a dataset corresponding to the first crop image and the crop breeding information specifically includes: The first crop image is preprocessed; the preprocessing includes noise reduction, contrast enhancement, and image segmentation. A set of crop features in the preprocessed first crop image is determined by a preset feature extraction algorithm; the set of crop features includes at least the following trait expression features: crop shape, crop texture, and crop color; The dataset is generated based on the key parts of the crop, the set of crop features, and the corresponding gene sequences in the first crop image pre-annotated by the user.

3. The crop breeding operation method based on a large model according to claim 1, characterized in that, The seed selection task originates from the user terminal wirelessly connected to the drone; the seed selection task includes a seed selection area and a seed selection path.

4. The crop breeding operation method based on a large model according to claim 1, characterized in that, The second crop image is identified by calling the trained seed selection model through a preset interface, specifically including: The seed selection model is used to determine the identification and breeding information corresponding to the second crop image; the seed selection model is a multimodal pre-trained model Unicoder-VL. The identified breeding information is matched with the target breeding information corresponding to the selection target; If the similarity between the identified breeding information and the target breeding information is greater than a preset threshold, the corresponding second crop image is determined to meet the selection target conditions; Otherwise, the selection criteria are not met.

5. The crop breeding operation method based on a large model according to claim 4, characterized in that, The method further includes: Once it is determined that the corresponding second crop image meets the seed selection target conditions, the acquisition location coordinates corresponding to the second crop image are determined; wherein, the acquisition location coordinates are sent by the UAV in real time; Based on the collected location coordinates, a corresponding transformation relationship tuple is determined; wherein, the transformation relationship tuple includes coefficients used to convert the coordinates of each pixel point of the second crop image into the coordinates of the corresponding cultivation base location; Determine the calibration pixel boxes in the second crop image that satisfy the selection target conditions; the calibration pixel boxes include one or more crops that satisfy the selection target conditions; Based on the coordinates of each pixel corresponding to the calibrated pixel box and the transformation relationship tuple, the corresponding set of location coordinates of the cultivation base is determined, which is the location information.

6. The crop breeding operation method based on a large model according to claim 1, characterized in that, The method further includes: When multiple drones perform the same seed selection task, a seed selection sub-path is determined for each drone in the seed selection area; the seed selection sub-paths are spliced ​​together to obtain the seed selection path corresponding to the seed selection task; the seed selection sub-path includes path nodes for capturing images of the second crop. The path node that is closest to the seed selection operation sub-path in each of the seed selection operation sub-paths is used as the splicing node; a seed selection operation sub-path includes one or two splicing nodes; The second crop images acquired by each of the stitching nodes are stitched together according to the seed selection operation path, and the corresponding stitched images are used as the second crop images.

7. A crop breeding operation device based on a large model, characterized in that, The device includes: At least one processor; and, A memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor to enable the at least one processor to perform a large-scale crop breeding operation method as described in any one of claims 1-6.

8. A non-volatile computer storage medium storing computer-executable instructions, characterized in that, The computer-executable instructions are capable of executing a crop breeding operation method based on a large model as described in any one of claims 1-6.

Images