Himalayan marmot hole intelligent identification model construction method fusing habitat information

Abstract

The embodiment of the invention discloses a Himalayan marmot hole intelligent identification model construction method fusing habitat information. The method comprises the steps of obtaining multiple unmanned aerial vehicle image maps of a Himalayan marmot hole; the habitat features of marmot holes and the surroundings of the marmot holes in the unmanned aerial vehicle images are marked, and the habitat features comprise hole earth mounds, vegetation, stones and animal traces; and training a target detection model by using each marked unmanned aerial vehicle image map, and using the trained model to improve the situation of marmot hole missed identification. The embodiment of the invention can overcome the missing detection condition of marmot holes.

Description

Technical Field

[0001] The embodiments of the present invention relate to the technical field of image target detection, and in particular to a method for constructing an intelligent recognition model of Himalayan marmot holes that integrates habitat information. Background Art

[0002] Himalayan marmot is the natural host of plague. By identifying Himalayan marmot burrows, we can understand the habitat and living conditions of Himalayan marmots and prevent the risk of plague.

[0003] Patent application CN112699852A provides a marmot intelligent identification and monitoring system. First, a set of marmot image sample libraries is built based on drone images, and a deep learning model is trained to obtain a marmot identification model. After that, a drone is used to take large-scale photos of the epidemic area to obtain images of marmots in the epidemic area, and the marmot density in the area and the potential plague risk are predicted. However, since marmots are active, restrictions on the drone shooting time and shooting range are added, which will cause some marmot objects in the image to be missed during identification. In addition, marmots are more sensitive to abnormal "invasion" objects such as drones and will escape the visual range of the drone. Combining the above two points, there is a large error in determining the marmot density through drone images, and the uncertainty in predicting the risk of plague is high.

[0004] Patent application CN114155973A provides a plague prediction method based on deep learning, combining drone images, unmanned vehicle data and a plague prediction model of the YOLO4 model. This patent collects hole data of small rodents such as yellow rats, Brandt's voles, and long-clawed gerbils, and trains the corresponding hole target detection model. According to this model, the location and number of small rodent holes directly exposed to the surface in the epidemic source can be obtained. On this basis, a variety of indicators are selected to construct a plague risk prediction model, monitor the changes in these indicators and combine the sampling vehicle data to obtain possible risk areas in the region. When selecting label data, this method directly marks the target objects that are obviously exposed in the drone image, so it can only identify the obvious holes in the image, and it is easy to miss objects blocked by vegetation, stones and other objects, and then there will be errors in the number of holes. Summary of the invention

[0005] An embodiment of the present invention provides a method for constructing an intelligent recognition model of Himalayan marmot holes to solve the problem of missed recognition of marmot holes.

[0006] In a first aspect, an embodiment of the present invention provides a method for constructing an intelligent identification model of Himalayan marmot holes integrating habitat information, comprising:

[0007] Get multiple drone images of Himalayan marmot caves;

[0008] Annotate the habitat features of the marmot holes and their surroundings in each drone image, where the habitat features include earth mounds at the hole entrance, vegetation, rocks, and animal traces;

[0009] The target detection model is trained using the annotated drone images, and the trained model is used to improve the identification of marmot hole leaks.

[0010] In a second aspect, an embodiment of the present invention provides an electronic device, the electronic device comprising:

[0011] one or more processors;

[0012] a memory for storing one or more programs,

[0013] When the one or more programs are executed by the one or more processors, the one or more processors implement the method for constructing an intelligent identification model of Himalayan marmot holes integrating habitat information as described in any embodiment.

[0014] In a third aspect, an embodiment of the present invention further provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the method for constructing an intelligent identification model of Himalayan marmot holes integrating habitat information as described in any embodiment.

[0015] In summary, this embodiment provides a method for constructing an intelligent recognition model of Himalayan marmot caves that integrates habitat information. The model assists in the investigation of natural plague foci by identifying Himalayan marmot caves rather than Himalayan marmots. First, a large-scale Himalayan marmot cave image is obtained by drone photography, and the iconic cave entrance features are defined together with the marmot cave as a generalized marmot cave, which is then applied as a labeling strategy to the Himalayan marmot cave training sample set and the recognition model construction method, reducing the risk of missed judgment of the target to be detected in the drone image, and improving the recognition ability of the target detection model for Himalayan marmot caves in a large range of situations. Compared with the narrow sense marmot caves directly exposed to the surface, the generalized marmot caves include not only marmot caves, but also habitat information such as cave entrance attachments and obstructions; applying it to the annotation and recognition of marmot caves, even if the target object is exposed in a small area in the image, the habitat information features near the cave entrance can still realize the recognition of marmot caves, effectively solving the model missed judgment problem caused by partial or complete occlusion of the target in the drone image. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] In order to more clearly illustrate the specific implementation methods of the present invention or the technical solutions in the prior art, the drawings required for use in the specific implementation methods or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are some implementation methods of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0017] Figure 1 It is a flow chart of a method for constructing an intelligent identification model of Himalayan marmot holes integrating habitat information provided by an embodiment of the present invention;

[0018] Figure 2 This is a diagram of the distribution pattern of marmot hole sizes provided by an embodiment of the present invention;

[0019] Figure 3 is a flowchart of another method for constructing an intelligent identification model of Himalayan marmot holes integrating habitat information provided by an embodiment of the present invention;

[0020] Figure 4 It is a schematic diagram of detection results of a broad sense marmot hole and a narrow sense marmot hole provided by an embodiment of the present invention;

[0021] Figure 5 A schematic diagram of the structure of an electronic device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0022] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the present invention will be described clearly and completely below. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work belong to the scope of protection of the present invention.

[0023] In the description of the present invention, it should be noted that the terms "center", "upper", "lower", "left", "right", "vertical", "horizontal", "inner", "outer", etc., indicating the orientation or positional relationship, are based on the orientation or positional relationship shown in the drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as limiting the present invention. In addition, the terms "first", "second", and "third" are used for descriptive purposes only, and cannot be understood as indicating or implying relative importance.

[0024] In the description of the present invention, it is also necessary to explain that, unless otherwise clearly specified and limited, the terms "installed", "connected", and "connected" should be understood in a broad sense, for example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection, or it can be indirectly connected through an intermediate medium, or it can be the internal communication of two components. For ordinary technicians in this field, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

[0025] Figure 1 is a flow chart of a method for constructing an intelligent identification model of Himalayan marmot holes integrating habitat information provided by an embodiment of the present invention. The method is executed by an electronic device, such as Figure 1 As shown, the specific steps include:

[0026] S110. Obtain multiple drone images of Himalayan marmot caves.

[0027] This embodiment constructs Himalayan marmot cave data, which is obtained by photographing with a drone. Considering the wide distribution range of Himalayan marmot caves, the drone can use a higher altitude and be equipped with a high-resolution camera to take orthophotos of the marmot caves, which include both the marmot cave objects and various habitat features around them.

[0028] S120. Label the habitat features of the marmot holes and their surroundings in each drone image.

[0029] This embodiment combines image analysis to obtain unique habitat features around the Himalayan marmot burrow from the image, such as the length of the burrow is generally wider than that of common pika burrows, and there are obvious mounds, slopes, and vegetation around the burrow, etc. These can be used as the iconic features that distinguish the Himalayan marmot burrow from other pika burrows.

[0030] Optionally, the size of the marmot hole can be determined based on the environmental features around the marmot hole and the features of the drone image, and the habitat features to be marked around the marmot hole can be selected from a large number of data sets. In a specific embodiment, the process can include the following steps:

[0031] Step 1. Determine the basic annotation size of the marmot holes based on the normal distribution of the sizes of the marmot holes in each drone image. This step determines the size of the annotation frame of the marmot holes based on the morphological characteristics of the hole. Specifically, the shapes of the marmot hole openings in the sampling area are mostly circular or elliptical. The direction of the marmot hole opening is used as the measurement direction of the length of the marmot hole opening, and the direction perpendicular to it is used as the measurement direction of the hole opening width. The measured length of the hole opening is between 10-74cm, and the width of the hole opening is between 8-70cm. The statistical results show that the average length of the marmot hole opening is 27.26cm, and the average width is 29.84cm. The data is normally distributed with the mean as the center, indicating that the size of the marmot holes in the survey area is highly consistent, such as Figure 2 As shown, the size of a labeling box that only considers marmot holes can be determined based on the statistical size situation.

[0032] Step 2: Based on the environment around the marmot hole and the features of the drone image, determine the habitat features that need to be marked around the marmot hole. Optionally, this embodiment provides three screening paths:

[0033] The first screening path determines the habitat features that need to be marked around the marmot holes based on the shape and size of the hole features, as well as the differences between the hole features and the surface and plants. Optionally, the significance of the shape and size of various hole features, as well as the differences from the surface and plants can be analyzed one by one, and finally the hole mounds are determined to be habitat features that need to be marked based on the fan-shaped or comet-shaped, large-size characteristics, significant color differences from the surface, and contrast with the exposed patches formed by plants. Specifically, the hole mounds are mostly fan-shaped or comet-shaped. Among the 159 marmot holes measured in the sampling area of ​​the Qinghai-Tibet Plateau, only 3 holes do not have obvious mounds. Nearly 98% of the holes have mounds around them. The surrounding mound slope features can be clearly seen in the drone images. The slopes of the mounds around the hole are also measured to be about 23-290 cm long and about 12-250 cm wide. The statistical results show that the average length and width of the mounds are 126.25cm and 126.04cm respectively, and the specific width varies with the type of marmot holes. In the bare land area, the color of the mounds is significantly different from the surrounding surface and is easy to identify; in the vegetation-covered area, the mounds cover the vegetation or destroy the vegetation, forming local exposed patches, which become an indirect sign of the hidden hole. Therefore, it can be used as a key sign to identify marmot holes.

[0034] The second screening path is to determine the habitat features that need to be marked around the marmot hole according to the hole selection and environmental adaptability of the marmot. Optionally, the importance of various hole characteristics in the site selection of marmots and their adaptability to the environment can be analyzed one by one, and finally, according to the shelters and soil characteristics that marmots prefer in the hole selection, the tree roots, stones, and earth mounds at the hole are determined as the habitat features that need to be marked. Specifically, the environmental adaptability of the hole selection: most marmot holes tend to choose areas with natural shelters, such as choosing under tree roots and stones, using shelters to enhance concealment and reduce exposure risks. In addition, it also has certain requirements for soil, such as giving priority to loose soil areas to reduce the difficulty of excavation. Optionally, according to the distribution of the hole and the surrounding earth mounds, the distribution of marmot holes can also include the following seven situations: large earth mounds, hole entrances along the roots of vegetation, hole entrances along stones, lush vegetation areas, dense stone areas, agricultural and pastoral transition zones, and mixed distribution with pika holes. These can all be used as marmot holes to mark and focus on capturing situations.

[0035] The third screening path is to determine the habitat features that need to be marked around the marmot hole based on the vertical shooting characteristics of the drone. Optionally, the obstructions in the vertical shooting of the drone can be analyzed one by one, and finally the obstructing elements such as vegetation, stones, and traces of animal activities at the hole entrance are used as habitat features that need to be marked. Specifically, there is a natural challenge in identifying marmot holes based on drone images: the vertical shooting characteristics of drone images lead to a low exposure rate of marmot holes, especially when the hole entrance is parallel or inclined to the ground, the hole entrance is obstructed by vegetation, stones, traces of animal activities, etc., which can easily form a "visual blind spot". Therefore, for the obstructed hole entrance, the obstruction itself can also be used as a landmark to identify the marmot hole.

[0036] For the convenience of distinction and description, in this embodiment, the habitat features that need to be marked obtained by the above three screening paths are respectively referred to as the first habitat feature, the second habitat feature and the third habitat feature. The three types of habitat features are taken as the union to form the habitat feature set that needs to be marked around the marmot hole. Ultimately, the set includes the earth mound at the hole entrance, vegetation, stones and animal traces. In this embodiment, these habitat features and the marmot hole together constitute the generalized marmot hole.

[0037] Step three, adjust the annotation size of the generalized marmot burrows according to the habitat characteristics. In subsequent operations, this application will annotate the generalized marmot burrows, so it is necessary to adjust the size of the initial annotation box to accommodate the above-mentioned habitat characteristics. The generalized marmot burrows can not only effectively deal with the obstruction of the burrow entrance, but also highlight the difference between marmot burrows and symbiotic animal burrows. For example, the length and width of the common pika burrow entrance are generally less than 10 cm, which is significantly smaller than the marmot burrows. This can be used as a basis for distinction. Even if marmot burrows and pika burrows coexist in some areas, the size difference is still a key feature for classification; for example, the definition of the generalized marmot burrows introduces fan-shaped or comet-shaped mounds, and their area is generally larger than 100cm 2, and there are vegetation or local exposed patches, stones and other features around it. Even if the entrance of the hole is completely blocked, its existence can still be inferred through associated habitat characteristics; in addition, the burrowing behavior of marmots is closely related to their living habits, which is essentially different from the common pika holes that are associated with slender mounds and no shelter.

[0038] Step 4: Determine the hierarchical structure of annotations based on the plan distribution map of marmot holes. This step takes into account the hierarchical distribution of marmot holes. By analyzing the plan distribution map of marmot holes, it is found that there is a distinction between main holes and satellite holes. The main holes have larger mounds and stronger functional cores; the satellite holes have smaller mounds and are distributed around the main holes to form a hierarchical network. Therefore, when annotating and sampling, the main holes and satellite holes should be sampled evenly to ensure that the data is representative when the model is trained later.

[0039] After the above preparations are completed, a labeling frame containing the marmot hole and its surrounding habitat features can be generated in each drone image; and the category, horizontal coordinate of the center point, vertical coordinate of the center point, width and height of the labeling frame can be stored. In a specific embodiment, the labeling process includes:

[0040] First, image screening is performed. Since target recognition is required later, it is necessary to select a high-quality dataset based on the above-mentioned drone image dataset. Specifically, when selecting data, combined with manual visual interpretation methods, images that do not contain the target object or have poor target quality are eliminated. At the same time, it is necessary to ensure that the habitat features determined in the above steps are included as much as possible around the cave entrance.

[0041] Then, data annotation and image cropping are performed. Based on the previous data collection, information acquisition and knowledge conversion, a generalized marmot hole is defined in this step. Different from the narrow marmot hole that only includes the hole entrance, the generalized marmot hole also includes the habitat characteristics around the hole entrance. Specifically, the content contained in a single annotation object is:

[0042]

[0043] in, represents an object labeled with a narrow sense marmot hole, represents an object labeled with a generalized marmot hole, Indicates the marmot hole object in the current annotation content. It represents the habitat characteristics around the target object, such as earth mounds, slopes, and vegetation, and k and i represent the number and index of the marked marmot holes, respectively.

[0044] Finally, the LabelImg tool is used for labeling, where the broad sense of marmot holes is defined as hole-mounds and the narrow sense of marmot holes is defined as holes. The YOLO format is used for storage, and its basic storage structure is as follows:

[0045]

[0046] in, Indicates the id of the current category. Indicates the x-coordinate of the center point of the annotation box. Indicates the y coordinate of the center point of the annotation box. and They represent the width and height of the annotation box respectively, and the following four parameters are the normalized result values.

[0047] In addition, since there is a lot of irrelevant information in the original image, it is necessary to crop the annotated image to reduce the time and space overhead of the model during training. Specifically, after the image is annotated, a 416×416 cropping area is drawn according to the range of the annotation box to crop the original image, so the final size of all images is 416×416.

[0048] S130, using each labeled drone image to train a target detection model, and the trained model is used to improve the identification of marmot hole leaks.

[0049] Optionally, data partitioning and model training are performed first. After a series of preprocessing operations are performed on the original image, the model training can be performed. In a specific embodiment, the image data set and the corresponding annotation data set can be divided into a training set and a test set as needed, wherein the training set is used to train the model parameters, and the parameters obtained by training are further improved in combination with the test set. Different partitioning ratios are used according to the characteristics of the data, and the partitioning ratio used here is 9:1.

[0050] Based on the divided data set, the transfer learning framework is used to assist model training. In general, transfer learning is a method of relearning new data set features based on an already created model to create a new model. Taking the YOLO11 model as an example, it has trained a variety of different models from small to large based on public data sets of different sizes, including n, s, m, l (corresponding to different model categories). Based on these basic models, transfer learning combined with marmot cave data can reduce the time to create recognition models.

[0051] In addition, the corresponding parameters need to be set before training the model. The core training parameters include image size (imgsz), training rounds (epochs), training batch size (batch), data enhancement strategy (close_mosaic), and the device used (device). The common enhancement operation is to randomly scale and crop the original image, and then use image stitching. In this way, the features of the target to be detected in the image can be enhanced. During specific training, according to the hardware conditions, imgsz is selected as 416, epochs is selected as 400, batch is selected as 8, close_mosaic is set to 20, the optimization strategy optimizer is set to stochastic gradient descent, and the device can be adjusted according to the computer equipment.

[0052] After training, the model can be evaluated. When judging the model effect, mAP50 is selected as the core indicator, which represents the degree of overlap between the predicted box and the true box. In addition, there are model precision and recall.

[0053]

[0054]

[0055] Among them, TP represents true positive, that is, samples predicted as positive and correctly predicted, FP represents samples predicted as positive but incorrectly predicted, and FN represents samples predicted as negative and correctly predicted. IoU represents intersection over union, where Intersection represents the intersection between the predicted box and the actual box, and Union represents the union between the predicted box and the actual box. mAP50 is the average of the average precision calculated under the condition that the IoU threshold is 0.5. The above method can also be used Figure 3 The flowchart shown is represented.

[0056] The effect of the model constructed in this embodiment is analyzed and verified below.

[0057] First, in theory, the labeling strategy based on generalized marmot holes will fully learn the habitat characteristics through neural networks in deep learning, and update the model parameters through gradient backpropagation:

[0058]

[0059] in, represents the parameters of the model during training, Represents the learning rate of parameter update during back propagation of the neural network, represents the auxiliary weight parameter, Habitat characteristics around the target object, represents the loss function based on the actual model parameters, It means partial derivative.

[0060] Based on this labeling strategy, the model will fully consider the habitat characteristics of the cave entrance when identifying the target. In this way, even if the model does not detect the actual cave entrance features, it can still calculate the surrounding areas of the cave entrance. The feature expression in the image can also identify the target object to a certain extent.

[0061] Secondly, in the experiment, the data set used in this embodiment was taken by a DJ1 Mavic 2 Pro drone. Since the Himalayan marmot holes are widely distributed, the flight altitude is set to 100 meters. The sampling area includes 29 typical natural plague foci located in the Tibet Autonomous Region. Through image sampling and annotation, 801 target objects were obtained. In order to ensure the balance of the sample, the hole and hole-mounds labeled objects are nearly half each. The hole-mounds label corresponds to the broad sense marmot hole mentioned in the present invention, and the hole label corresponds to the narrow sense marmot hole.

[0062] During the model training process, since YOLO8 and YOLO11 are currently widely used in target detection tasks and have better recognition accuracy than other target detection models, the YOLO8n and YOLO11n models are used to detect holes and hole-mounds to determine the recognition accuracy of different models for these two types of objects. In the setting of training parameters, in order to control the images of other factors, the same values ​​are set for the core training parameters, where imgsz is selected as 416, epochs is selected as 400, batch is selected as 8, close_mosaic is set to 20, and the optimization strategy optimizer is set to stochastic gradient descent, and the device is selected as cpu. The final training results are shown in the following table:

[0063] Model YOLO8n YOLO11n mAP50 0.924 0.929 mAP50-95 0.490 0.443 hole(mAP50) 0.902 0.861 hole-mounds(mAP50) 0.935 0.876 precision 0.918 0.935 recall 0.803 0.831

[0064] As can be seen from the table, by using hole-mounds for annotation, it is easier to identify in the picture because its features are richer and more unique. Whether using YOLO8n or YOLO11n, its accuracy has been improved to a certain extent, by 3.59% and 1.74% respectively.

[0065] From this, it can be seen that the use of a fusion habitat feature labeling strategy can improve the recognition accuracy of Himalayan marmot burrows and reduce the missed detection of target objects, proving that the strategy is effective. Figure 4 This is the target detection result obtained through experiments.

[0066] In summary, this embodiment provides a method for constructing an intelligent identification model of Himalayan marmot caves that integrates habitat information. The model assists in the investigation of natural plague foci by identifying Himalayan marmot caves rather than Himalayan marmots. First, a large-scale image of Himalayan marmot caves is obtained by drone photography, and the habitat characteristics of various cave entrances are analyzed from the perspectives of feature size, morphology, contrast, marmot cave site selection, and obstructions. The iconic information such as mounds, stones, and vegetation at the cave entrances are screened out, and it is defined together with the marmot caves as a generalized marmot cave, and then it is applied as a labeling strategy to the Himalayan marmot cave training sample set and the recognition model construction method, which reduces the risk of missing the target to be detected in the drone image and improves the target detection model's recognition ability for Himalayan marmot caves in a large range of situations. Compared with the narrow sense of marmot burrows that are directly exposed to the surface, the broad sense of marmot burrows includes not only marmot burrows, but also habitat information such as burrow attachments and obstructions. Applying it to the annotation and identification of marmot burrows, even if the target object is exposed in a small area in the image, the habitat information characteristics near the burrow entrance can still be used to identify the marmot burrow, effectively solving the problem of model missed judgment caused by partial or complete occlusion of the target in UAV images.

[0067] Figure 5 A schematic diagram of the structure of an electronic device provided by an embodiment of the present invention is shown in FIG. Figure 5 As shown, the device includes a processor 60, a memory 61, an input device 62 and an output device 63; the number of processors 60 in the device can be one or more. Figure 5 A processor 60 is taken as an example; the processor 60, the memory 61, the input device 62 and the output device 63 in the device can be connected by a bus or other means. Figure 5 The example of connecting through bus is taken in the following.

[0068] The memory 61 is a computer-readable storage medium that can be used to store software programs, computer executable programs and modules, such as program instructions / modules corresponding to the method for constructing an intelligent identification model of a Himalayan marmot hole integrating habitat information in the embodiment of the present invention. The processor 60 executes various functional applications and data processing of the device by running the software programs, instructions and modules stored in the memory 61, that is, realizing the above-mentioned method for constructing an intelligent identification model of a Himalayan marmot hole integrating habitat information.

[0069] The memory 61 may mainly include a program storage area and a data storage area, wherein the program storage area may store an operating system and at least one application required for a function; the data storage area may store data created according to the use of the terminal, etc. In addition, the memory 61 may include a high-speed random access memory, and may also include a non-volatile memory, such as at least one disk storage device, a flash memory device, or other non-volatile solid-state storage device. In some instances, the memory 61 may further include a memory remotely arranged relative to the processor 60, and these remote memories may be connected to the device via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0070] The input device 62 may be used to receive input digital or character information and generate key signal input related to user settings and function control of the device. The output device 63 may include a display device such as a display screen.

[0071] An embodiment of the present invention further provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the method for constructing an intelligent identification model of Himalayan marmot holes integrating habitat information of any embodiment.

[0072] The computer storage medium of the embodiment of the present invention can adopt any combination of one or more computer-readable media. The computer-readable medium can be a computer-readable signal medium or a computer-readable storage medium. The computer-readable storage medium can be, for example, but not limited to, a system, device or device of electricity, magnetism, light, electromagnetic, infrared, or semiconductor, or any combination of the above. More specific examples (non-exhaustive list) of computer-readable storage media include: an electrical connection with one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In this document, a computer-readable storage medium can be any tangible medium containing or storing a program, which can be used by an instruction execution system, device or device or used in combination with it.

[0073] Computer-readable signal media may include data signals propagated in baseband or as part of a carrier wave, which carry computer-readable program code. Such propagated data signals may take a variety of forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination of the above. Computer-readable signal media may also be any computer-readable medium other than a computer-readable storage medium, which may send, propagate, or transmit a program for use by or in conjunction with an instruction execution system, apparatus, or device.

[0074] The program code embodied on the computer readable medium may be transmitted using any appropriate medium, including but not limited to wireless, wireline, optical fiber cable, RF, etc., or any suitable combination of the foregoing.

[0075] Computer program code for performing the operation of the present invention may be written in one or more programming languages ​​or a combination thereof, including object-oriented programming languages ​​such as Java, Smalltalk, C++, and conventional procedural programming languages ​​such as C or similar programming languages. The program code may be executed entirely on the user's computer, partially on the user's computer, as a separate software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In the case of a remote computer, the remote computer may be connected to the user's computer through any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0076] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or replace some or all of the technical features therein by equivalents. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the technical solutions of the embodiments of the present invention.

Claims

1. A method for constructing an intelligent identification model for Himalayan marmot holes integrating habitat information, characterized in that: include: Get multiple drone images of Himalayan marmot caves; Annotate the habitat features of the marmot holes and their surroundings in each drone image, where the habitat features include earth mounds at the hole entrance, vegetation, rocks, and animal traces; The target detection model is trained using the annotated drone images, and the trained model is used to improve the identification of marmot hole leaks.

2. The method according to claim 1, characterized in that The marking of the marmot holes and the surrounding habitat features in each drone image includes: Based on the environment around the marmot burrows and the features of drone images, the marking dimensions of the marmot burrows and the habitat features that need to be marked around the marmot burrows are determined.

3. The method according to claim 2, characterized in that The marking size of the marmot hole and the habitat features to be marked around the marmot hole are determined based on the environment around the marmot hole and the features of the drone image, including: Based on the normal distribution of the sizes of marmot holes in each UAV image, the basic dimensions of the marmot holes were determined; Based on the environment around the marmot hole and the characteristics of drone images, determine the habitat features that need to be marked around the marmot hole; The characteristics of each habitat are used to adjust the labeled dimensions of the generalized marmot holes; Determine the labeling hierarchy based on the plan distribution map of marmot holes.

4. The method according to claim 3, characterized in that Based on the environment around the marmot hole and the features of the drone image, the habitat features that need to be marked around the marmot hole are determined, including: Determine the first habitat feature to be marked around the marmot hole based on the shape and size of the hole feature, as well as the difference between the hole feature and the ground surface and plants; According to the location of the marmot's hole and its environmental adaptability, determine the secondary habitat features that need to be marked around the marmot hole; Based on the vertical shooting characteristics of the drone, the third habitat features that need to be marked around the marmot holes are determined; The first habitat feature, the second habitat feature and the third habitat feature are taken as a union to form a set of habitat features that need to be marked around the marmot hole.

5. The method according to claim 4, characterized in that Determining the first habitat feature to be marked around the marmot hole based on the shape and size of the hole entrance feature, and the difference between the hole entrance feature and the ground surface and plants, includes: determining the hole entrance mound as the first habitat feature to be marked based on the fan-shaped or comet-shaped, large-size characteristics, significant color difference from the ground surface, and comparison with exposed patches formed by plants; The second habitat feature to be marked around the marmot hole is determined according to the hole site selection and environmental adaptability of the marmot hole, including: determining the tree roots, stones, and earth mounds at the hole as the second habitat feature to be marked according to the shelter and soil characteristics that the marmot prefers in the hole site selection; The method of determining the third habitat feature to be marked around the marmot hole according to the vertical shooting characteristics of the drone includes: determining the vegetation, stones, and animal activity traces at the hole entrance as the third habitat feature to be marked according to the obstructions in the vertical shooting of the drone.

6. The method according to claim 3, characterized in that Determining the labeled hierarchical structure according to the plan distribution map of the marmot holes includes: Extract the hierarchical distribution of marmot holes from the planar distribution map of marmot holes; According to the hierarchical distribution, the types of marmot holes to be labeled are determined, and a labeling hierarchy for uniformly sampling different types of marmot holes is determined.

7. The method according to claim 1, characterized in that The marking of the marmot holes and the surrounding habitat features in each drone image includes: In each drone image, a label box is generated to contain the habitat features of the marmot burrow and its surroundings; Stores the category, center point horizontal coordinate, center point vertical coordinate, width and height of the annotation box.

8. The method according to claim 1, characterized in that The target detection model adopts the YOLO model.

9. An electronic device, characterized in that: include: one or more processors; a memory for storing one or more programs, When the one or more programs are executed by the one or more processors, the one or more processors implement the method for constructing an intelligent identification model of Himalayan marmot holes integrating habitat information as described in any one of claims 1-8.

10. A computer-readable storage medium, characterized in that: A computer program is stored thereon, which, when executed by a processor, implements the method for constructing an intelligent identification model of Himalayan marmot holes integrating habitat information as described in any one of claims 1-8.

Images