A method for investigating Chinese tachypleus larvae based on remote sensing technology and image intelligent recognition

Abstract

This invention discloses a method for surveying juvenile horseshoe crabs in China based on remote sensing technology and intelligent image recognition. Through a process of rapid comprehensive survey and precise regional survey, a drone equipped with a lightweight YOLO recognition model, RTK positioning, and laser ranging is used to conduct a comprehensive survey of potential habitats, identify juvenile distribution hotspots, and delineate precise survey areas. Then, a full-frame mapping camera is used to conduct ultra-high-resolution drone photogrammetry, combined with an instance segmentation model to complete juvenile contour extraction, coordinate positioning, density statistics, and age determination. This invention overcomes the shortcomings of traditional foot surveys, such as tidal limitations, small coverage area, susceptibility to environmental interference, high missed detection rate, and high personnel risk. It achieves non-contact, high-efficiency, high-precision, and low-disturbance juvenile horseshoe crab surveys, providing a scientific and efficient technical solution for the monitoring of juvenile horseshoe crabs in China and the protection of endangered species.

Description

Technical Field

[0001] This invention relates to the field of image analysis and biological surveys, and more specifically, to a method for surveying Chinese horseshoe crab larvae based on remote sensing technology and intelligent image recognition. Background Technology

[0002] Chinese horseshoe crab ( Tachypleus tridentatus Horseshoe crabs (Limulus hornulatus) are marine benthic animals belonging to the phylum Arthropoda, class Mesostomata, order Limulus, family Limulidae, and genus Limulus. Because they have retained many primitive morphological features, they are known as "living fossils" and are of great significance in the study of life evolution. This species has valuable biomedical value; horseshoe crab reagents made from their blood are the optimal method for detecting bacterial endotoxins in the pharmaceutical, food, and cosmetic industries.

[0003] Horseshoe crab larvae inhabit the intertidal mudflats, a crucial stage for population replenishment and continuation. Field surveys can reveal their population size, distribution, and habitat characteristics, providing vital information for assessing population status, identifying key nurseries, and developing conservation measures. Current survey methods primarily involve manual trekking along transects and quadrats, but this approach has several drawbacks: it is limited by tidal timing, resulting in short working windows and hindering continuous surveys on a large spatial scale; the high randomness of sampling may lead to insufficient data representativeness, affecting the accuracy and reliability of results; the survey process disturbs the intertidal substrate structure, impacting larval life; and survey personnel face risks such as being trapped by tidal currents, getting stuck in mudflats, and being attacked by toxic or aggressive marine life. Therefore, there is an urgent need to establish an efficient, accurate, and low-disturbance survey method.

[0004] The integration of ultra-high resolution UAV photogrammetry, remote sensing analysis, and intelligent image recognition technology has provided a breakthrough solution for the survey of horseshoe crab larvae in China. UAVs can complete image acquisition of large areas of mudflats during the short tidal window, greatly expanding the coverage of a single survey and enabling continuous and comprehensive monitoring of the entire habitat. UAV remote sensing, as a non-contact survey method, avoids interference with the intertidal substrate structure and horseshoe crab larvae, more realistically reflecting their distribution and behavioral patterns in their natural state, while also ensuring the safety of survey personnel. Ultra-high resolution imagery can depict sub-millimeter-level details, overcoming the challenges of the larvae's small size and high similarity to the mudflat color, significantly reducing the missed detection rate. Meanwhile, intelligent image recognition technology can efficiently and accurately analyze these massive amounts of high-definition image data, automatically identifying and counting the number of larvae, reducing human error and further improving data quality. Summary of the Invention

[0005] This invention overcomes the shortcomings of existing technologies and proposes a survey method for juvenile Chinese horseshoe crabs based on remote sensing technology and intelligent image recognition. Specifically, it is a survey method for juvenile Chinese horseshoe crabs based on ultra-high resolution UAV remote sensing technology photogrammetry and intelligent image recognition. Compared with traditional ground survey methods, it has higher efficiency and accuracy, avoids interference with the ecological environment, ensures the personal safety of survey personnel, and can provide technical support for the precise protection and scientific management of juvenile Chinese horseshoe crabs.

[0006] The first aspect of this invention provides a method for surveying juvenile horseshoe crabs in China based on remote sensing technology and intelligent image recognition, comprising: S1: Using a mapping drone, navigate and monitor within a preset area based on a preset path, and collect image data and location information in real time; S2: Target detection is performed on real-time acquired image data using a target detection algorithm. When a juvenile Chinese horseshoe crab is identified, the mapping drone performs a close-up operation on the target. Using the RTK positioning module and the synchronously acquired multi-dimensional location information, a spatial coordinate transformation algorithm is introduced to calculate the accurate geographic coordinates and upload the identification data to the cloud server. S3: Based on the identification data, conduct spatial distribution trend analysis of Chinese horseshoe crab larvae, and select the survey range and corresponding boundary coordinate data based on the density distribution; S4: Set the flight path and parameters of the survey drone according to the survey range. During the survey, take photos at preset time intervals. At the same time, based on the built-in RTK positioning module and the camera's exposure time synchronization mechanism, calculate the high-precision geographic coordinates of the center point of each photo and embed the coordinates into the photo. Then, process the photos to generate an orthophoto map. S5: Import the orthophoto image into the instance segmentation model to identify Chinese horseshoe crab juveniles and extract pixel-level contour features. Combine the spatial information of the orthophoto image to construct an affine transformation matrix and calculate the geographic coordinates of the inflection points in the pixel-level contour features to form a coordinate vector file. S6: Import the coordinate vector file into the geographic information system to assess the population distribution density, age, and spatial aggregation characteristics of Chinese horseshoe crab larvae.

[0007] In this solution, S1 specifically refers to: The mapping UAV includes a visual acquisition module, a spatial positioning and ranging module, a high-speed wireless communication module, a remote sensing spectroscopic device, and a built-in high-performance computing unit; The visual acquisition module includes a visible light fixed-focus camera and a zoom camera. The visible light fixed-focus camera is used to continuously acquire video of the mudflat surface during the drone's flight, and the zoom camera is used to take close-up photos of the identified horseshoe crab larvae. The built-in high-performance computing unit establishes a communication connection with the visual acquisition module. It is equipped with a juvenile horseshoe crab intelligent recognition model and is configured to receive image data from the visible light fixed-focus camera in real time and perform target inference and detection. When a target is detected, it controls the zoom camera to move.

[0008] In this solution, S1 further includes: The laser rangefinder has a ranging accuracy of ±1 cm and a working range of 3 m to 3000 m; the visible light fixed-focus camera has a focal length set to 24 mm and an effective pixel count of 20 million; the zoom camera has a variable focal length range set to 30 mm to 800 mm and an effective pixel count of 20 million.

[0009] In this solution, S1 further includes: Based on the preset area, the bow-shaped flight path and preset flight parameters are set for navigation and monitoring. Specifically, the flight altitude relative to the takeoff point is set to 6 m, the flight speed is set to 5 m / s, the heading overlap rate is set to 60%, and the lateral overlap rate is set to 50%. During navigation and monitoring, the visual acquisition module of the mapping UAV continuously acquires surface image streams of the habitat mudflats at a preset frame rate; Using a built-in high-performance computing unit, the system synchronously receives surface image streams and runs a pre-loaded lightweight Chinese horseshoe crab larvae recognition model in real time to perform frame-by-frame inference detection on the image stream.

[0010] In this solution, S2 specifically refers to: The model identifies and detects juvenile horseshoe crabs in real time. If the confidence level is greater than or equal to 60%, a remote sensing spectral device is used to obtain spectral images for secondary confirmation of the object. The zoom camera is then triggered to automatically zoom. The relative position of the target in the field of view is calculated by using the two-dimensional bounding box coordinates of the target object output by the model. The gimbal is then driven to adjust its attitude and the zoom camera is controlled to automatically zoom, so that the juvenile horseshoe crab is centered and reaches the preset magnification ratio. Subsequently, the zoom camera is controlled to take a close-up image of the target object. While the zoom camera is taking the close-up image, the computing unit calls the RTK positioning module to obtain the reference geographic coordinates of the current observation device and simultaneously triggers the laser rangefinder to obtain the straight-line distance and spatial pitch and azimuth angles from the observation device to the juvenile horseshoe crab target. Based on the reference geographic coordinates, straight-line distance, spatial pitch angle and azimuth angle, the precise geographic coordinates of the Chinese horseshoe crab juvenile target are calculated by a spatial coordinate transformation algorithm, and the precise geographic coordinates are converted into metadata and embedded in the Exif information layer of the close-up photo of the target object; The close-up photo and its corresponding precise geographic coordinates are uploaded to a remote cloud server via an onboard high-speed wireless communication module.

[0011] In this scheme, the spatial coordinate transformation algorithm in S2 specifically includes: A local coordinate system is established with the antenna phase center of the RTK positioning module as the origin. Combined with the attitude data obtained by the IMU inertial measurement unit built into the mapping UAV, the polar coordinate system data measured by the laser rangefinder is converted into relative spatial rectangular coordinates. Through the CGCS2000 coordinate system transformation model, the relative spatial rectangular coordinates are superimposed on the reference geographic coordinates to obtain the precise geographic coordinates of the Chinese horseshoe crab juvenile containing longitude, latitude and elevation information.

[0012] In this solution, S3 specifically refers to: Once the mapping UAV has completed its flight mission, the cloud server calls upon the integrated geographic information system module to perform kernel density analysis and spatial hotspot analysis on the obtained distribution points of Chinese horseshoe crab larvae based on precise geographic coordinates, in order to quantitatively assess the spatial distribution trend of Chinese horseshoe crab larvae. Based on the spatial distribution trend of horseshoe crab larvae in China, the cloud server automatically typeset and generates a survey report, which includes three types of visualization maps: a distribution map of horseshoe crab larvae in China, a nuclear density map of horseshoe crab larvae in China, and a distribution heat map of horseshoe crab larvae in China. In the results of the kernel density analysis, the spatial region with a kernel density value greater than or equal to 3 ind / 100 m² is extracted. The spatial region is defined as the precise survey range of Chinese horseshoe crab larvae. The edge contour of the range is vectorized and the boundary coordinate data of the precise survey range is output.

[0013] In this solution, S4 specifically refers to: The survey drone includes a built-in RTK positioning module and a full-frame mapping camera with a resolution of 60 megapixels or more and a focal length of 40 mm or more. The boundary coordinates of the precise survey area are imported into the UAV flight path planning software to generate the photogrammetric survey flight path and parameters. The specific flight path and parameters include: the flight altitude relative to the take-off point is set to 8 m, the flight speed is set to 1 m / s, the forward overlap rate is set to 60%, the lateral overlap rate is set to 50%, and the camera pitch angle is -90°. The survey drone is controlled to perform survey tasks according to the flight path. During the survey tasks, the survey drone takes aerial photos at a 1-second interval. At each camera exposure moment, the timestamp of the exposure moment is obtained through the built-in RTK positioning module and the camera's exposure time synchronization mechanism, and the RTK antenna phase center coordinates of the UAV system at each timestamp are recorded. Based on the preset spatial offset between the RTK antenna phase center and the camera optical center, the eccentricity compensation calculation is performed on the coordinates of the RTK antenna phase center to obtain the high-precision geographic coordinates of the camera center point at the time of the corresponding orthophoto, and then written into the Exif information of the orthophoto. After the drone flight path is completed, all aerial photographs are imported into photogrammetric data analysis software. The software first extracts image feature points from overlapping areas of adjacent aerial photographs and performs corresponding point matching. Using the high-precision geographic coordinates in the Exif information of the aerial photographs as initial position constraints, aerial triangulation is performed to obtain the precise exterior orientation elements of each aerial photograph. Based on the calculated exterior orientation elements and matching feature points, a dense point cloud is generated, and a digital surface model is constructed. The digital surface model is then used to orthorectify and mosaic the aerial photographs to generate an orthophoto map covering the precise survey area.

[0014] In this solution, S5 includes: The orthophoto image is input into a pre-trained segmentation model for juvenile Chinese horseshoe crabs; The orthophoto image is analyzed using an instance segmentation model to identify juvenile Chinese horseshoe crabs in the orthophoto image and extract the contours of each juvenile Chinese horseshoe crab; the pixel coordinates of each inflection point of the contour are calculated. Read the spatial reference information of the digital orthophoto image, wherein the spatial reference information includes at least the origin geographic coordinates and the ground sampling distance parameter; Based on the spatial reference information, an affine transformation matrix is ​​constructed from the pixel coordinate system to the real geographic coordinate system; The pixel coordinates of the contour inflection points are input into the affine transformation matrix for coordinate mapping calculation to obtain the geographic coordinates of the contour inflection points of the Chinese horseshoe crab larvae. The geographic coordinates were converted into the outline information of the Chinese horseshoe crab larvae and saved as a vector file in shp format.

[0015] In this solution, S6 specifically refers to: Importing the vector file of the juvenile Chinese horseshoe crab into a geographic information system for quantitative analysis, the quantitative analysis including at least: The number of horseshoe crab larvae in the survey area was counted, and the population density of horseshoe crab larvae was calculated based on the area parameter of the survey area. Extract the polygonal geometric features of the Chinese horseshoe crab outline, calculate the cephalothorax width of each Chinese horseshoe crab larva based on a preset measurement baseline, and infer the age of the Chinese horseshoe crab larvae based on the cephalothorax width; Spatial overlay analysis was performed on the vector profile file of Chinese horseshoe crab larvae and the distribution map of environmental factors in the survey area. Spatial clustering algorithm was used to extract the aggregation characteristics of the spatial distribution of the Chinese horseshoe crab larvae, and a spatial distribution correlation pattern between the aggregation characteristics and environmental factors such as substrate type and tidal channel distribution was established.

[0016] The second aspect of the present invention also provides a survey system for juvenile horseshoe crabs in China based on remote sensing technology and intelligent image recognition. The system includes: a mapping quadcopter UAV platform, a survey UAV platform, a cloud server, and a communication module. When the system is running, it implements all steps S1-S6 of the above-described survey method for juvenile horseshoe crabs in China based on remote sensing technology and intelligent image recognition.

[0017] This invention achieves the following beneficial effects by introducing ultra-high resolution UAV photogrammetry technology, remote sensing analysis, and intelligent image recognition technology: Significantly improves the efficiency and spatial coverage of horseshoe crab larvae surveys in China. The use of drones enables the collection of images of large areas of mudflats within a short tidal window, achieving continuous and comprehensive monitoring of horseshoe crab larvae habitats in China. Ultra-high resolution UAV remote sensing technology, as a non-contact survey method, avoids direct interference with the intertidal bottom structure and juvenile horseshoe crabs, and can more realistically reflect their distribution and behavior patterns in the natural state; moreover, remote sensing images can depict sub-millimeter level details, overcoming the problem that juvenile horseshoe crabs are small in size and highly similar in color to the mudflats, and significantly reducing the missed detection rate. By using intelligent image recognition technology to efficiently and accurately analyze massive amounts of image data, the number of juveniles can be automatically identified and counted, thereby reducing human error, improving data quality and reliability, and avoiding various survey safety risks for investigators. Attached Figure Description

[0018] Figure 1 A flowchart of a method for surveying juvenile horseshoe crabs in China based on remote sensing technology and intelligent image recognition according to the present invention is shown. Figure 2 A block diagram of a Chinese horseshoe crab juvenile survey system based on remote sensing technology and intelligent image recognition is shown. Detailed Implementation

[0019] To better understand the above-mentioned objectives, features, and advantages of the present invention, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. It should be noted that, unless otherwise specified, the embodiments and features described in these embodiments can be combined with each other.

[0020] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and therefore the scope of protection of the invention is not limited to the specific embodiments disclosed below.

[0021] Figure 1 The flowchart of a method for surveying Chinese horseshoe crab larvae based on remote sensing technology and intelligent image recognition according to the present invention is shown.

[0022] like Figure 1 As shown, the first aspect of this invention provides a method for surveying juvenile horseshoe crabs in China based on remote sensing technology and intelligent image recognition, comprising: S1: Using a mapping drone, navigate and monitor within a preset area based on a preset path, and collect image data and location information in real time; S2: Target detection is performed on real-time acquired image data through target detection algorithm. When the juvenile Chinese horseshoe crab is identified, the mapping drone performs a close-up operation on the target. The RTK positioning module and the multi-dimensional location information acquired synchronously are used to introduce a spatial coordinate transformation algorithm to calculate the accurate geographic coordinates and upload the identification data to the cloud server. S3: Based on the identification data, conduct spatial distribution trend analysis of Chinese horseshoe crab larvae, and select the survey range and corresponding boundary coordinate data based on the density distribution; S4: Set the flight path and parameters of the survey drone according to the survey range. During the survey, take photos at preset time intervals. At the same time, based on the built-in RTK positioning module and the camera's exposure time synchronization mechanism, calculate the high-precision geographic coordinates of the center point of each photo and embed the coordinates into the photo. Then, process the photos to generate orthophotos. S5: Import the orthophoto image into the instance segmentation model to identify Chinese horseshoe crab juveniles and extract pixel-level contour features. Combine the spatial information of the orthophoto image to construct an affine transformation matrix and calculate the geographic coordinates of the inflection points in the pixel-level contour features to form a coordinate vector file. S6: Import the coordinate vector file into the geographic information system to assess the population distribution density, age, and spatial aggregation characteristics of Chinese horseshoe crab larvae.

[0023] According to an embodiment of the present invention, S1 specifically includes: The mapping UAV includes a visual acquisition module, a spatial positioning and ranging module, a high-speed wireless communication module, a remote sensing spectroscopic device, and a built-in high-performance computing unit; The visual acquisition module includes a visible light fixed-focus camera and a zoom camera. The visible light fixed-focus camera is used to continuously acquire video of the mudflat surface during the drone's flight, and the zoom camera is used to take close-up photos of the identified horseshoe crab larvae. The built-in high-performance computing unit establishes a communication connection with the visual acquisition module. It is equipped with a juvenile horseshoe crab intelligent recognition model and is configured to receive image data from the visible light fixed-focus camera in real time and perform target inference and detection. When a target is detected, it controls the zoom camera to move.

[0024] It should be noted that the surveying process is carried out through a surveying quadcopter drone platform, which includes the surveying drone.

[0025] The spatial positioning and ranging module includes a high-precision real-time dynamic RTK positioning module and a high-precision laser rangefinder, both of which are electrically connected to the built-in high-performance computing unit. They are used to simultaneously measure when taking close-up photos in order to accurately obtain the spatial geographic coordinates of the Chinese horseshoe crab juveniles.

[0026] According to an embodiment of the present invention, step S1 further includes: The high-precision laser rangefinder has a ranging accuracy of ±1 cm and a working range of 3 m to 3000 m; the visible light fixed-focus camera has a focal length set to 24 mm and an effective pixel count of 20 million; the zoom camera has a variable focal length range set to 30 mm to 800 mm and an effective pixel count of 20 million.

[0027] It should be noted that the above parameters are those of the preferred embodiment. Depending on the research needs, different monitoring equipment or camera parameters may be replaced to ensure the adaptability of the surveying and mapping in different environments.

[0028] According to an embodiment of the present invention, step S1 further includes: Based on the preset area, the bow-shaped flight path and preset flight parameters are set for navigation and monitoring. Specifically, the flight altitude relative to the takeoff point is set to 6 m, the flight speed is set to 5 m / s, the heading overlap rate is set to 60%, and the lateral overlap rate is set to 50%. During navigation and monitoring, the visual acquisition module of the mapping UAV continuously acquires surface image streams of the habitat mudflats at a preset frame rate; Using a built-in high-performance computing unit, the system synchronously receives surface image streams and runs a pre-loaded lightweight Chinese horseshoe crab larvae recognition model in real time to perform frame-by-frame inference detection on the image stream.

[0029] This can be understood as follows: Pre-defined coordinate data of potential habitat ranges for juvenile Chinese horseshoe crabs (i.e., preset area information) is imported into UAV flight path planning software. The software then generates an arc-shaped flight path covering the potential habitat based on this coordinate data. During UAV flight, a pre-loaded lightweight Chinese horseshoe crab juvenile recognition model is run in real-time to perform frame-by-frame inference detection on the image stream, enabling real-time identification of juvenile Chinese horseshoe crab targets within the field of view. This lightweight Chinese horseshoe crab juvenile recognition model is generated by optimizing and compressing a multi-scale feature dataset of Chinese horseshoe crab juveniles, based on the YOLO target detection algorithm as its foundation network.

[0030] According to an embodiment of the present invention, step S2 specifically includes: The model identifies and detects juvenile horseshoe crabs in real time. If the confidence level is greater than or equal to 60%, a remote sensing spectral device is used to obtain spectral images for secondary confirmation of the object. The zoom camera is then triggered to automatically zoom. The relative position of the target in the field of view is calculated by using the two-dimensional bounding box coordinates of the target object output by the model. The gimbal is then driven to adjust its attitude and the zoom camera is controlled to automatically zoom, so that the juvenile horseshoe crab is centered and reaches the preset magnification ratio. Subsequently, the zoom camera is controlled to take a close-up image of the target object. While the zoom camera is taking the close-up image, the computing unit calls the RTK positioning module to obtain the reference geographic coordinates of the current observation device and simultaneously triggers the laser rangefinder to obtain the straight-line distance and spatial pitch and azimuth angles from the observation device to the juvenile horseshoe crab target. Based on the reference geographic coordinates, straight-line distance, spatial pitch angle and azimuth angle, the precise geographic coordinates of the Chinese horseshoe crab juvenile target are calculated by a spatial coordinate transformation algorithm, and the precise geographic coordinates are converted into metadata and embedded in the Exif information layer of the close-up photo of the target object; The close-up photo and its corresponding precise geographic coordinates are uploaded to a remote cloud server via an onboard high-speed wireless communication module.

[0031] Here, it can be understood that the target detection algorithm is implemented through a recognition model. The current observation device is the current mapping UAV. Data is uploaded to a remote cloud server for real-time visualization of the spatial distribution of juvenile horseshoe crabs. Introducing remote sensing spectroscopic equipment to acquire spectral images for secondary object identification is an optional process, used for refined discrimination using remote sensing imaging.

[0032] According to an embodiment of the present invention, in step S2, the spatial coordinate transformation algorithm specifically includes: A local coordinate system is established with the antenna phase center of the RTK positioning module as the origin. Combined with the attitude data obtained by the IMU inertial measurement unit built into the mapping UAV, the polar coordinate system data measured by the laser rangefinder is converted into relative spatial rectangular coordinates. Through the CGCS2000 coordinate system transformation model, the relative spatial rectangular coordinates are superimposed on the reference geographic coordinates to obtain the precise geographic coordinates of the Chinese horseshoe crab juvenile containing longitude, latitude and elevation information.

[0033] According to an embodiment of the present invention, the recognition model specifically includes: Images of juvenile horseshoe crabs under different lighting conditions and in different mudflat environments were collected. Considering the small size of the juvenile horseshoe crabs and the fact that they are generally covered with mud film, Mosaic stitching, random cropping, and adaptive histogram equalization were introduced to enhance the images. Gaussian noise was added, the image dataset was labeled, and training and testing sets were generated. Based on the YOLOv8 algorithm, the backbone feature extraction network of the original framework was replaced with the lightweight network MobileNetV3, and the attention mechanism module CBAM was introduced into the feature fusion layer of the network neck to build a recognition model. The training set is input into the recognition model for iterative training. The learning rate, batch size and number of iterations are initialized, and CIoU is used as the bounding box regression loss function. During the training process, the learning rate is dynamically adjusted according to the loss value of the validation set until the model loss function converges, and the initial recognition model is obtained.

[0034] The initial recognition model is pruned to remove redundant and minimally weighted network connections. Post-training quantization is used to convert floating-point parameters in the model into integer parameters, generating the final lightweight Chinese horseshoe crab juvenile recognition model. The recognition model is then compiled into TensorRT format adapted for airborne computing units and loaded.

[0035] It should be noted that, considering the small size and mud-covered bodies of juvenile horseshoe crabs, data augmentation was performed on the original images. This data augmentation included, but was not limited to, Mosaic stitching, random cropping, adaptive histogram equalization, and adding Gaussian noise to increase sample diversity. Subsequently, the juvenile horseshoe crabs in the images were manually labeled using an annotation tool to generate training and testing sets. The recognition model was specifically a lightweight YOLO network model. Here, the YOLOv8 algorithm was used as the basic framework, with lightweight improvements. The backbone network of the original framework was replaced with a lightweight network (MobileNet) to reduce the number of parameters and floating-point operations (FLOPs). Simultaneously, a focus mechanism module (CBAM) was introduced into the neck feature fusion layer to enhance the model's weight allocation for the contours and texture features of small targets against the complex mudflat background, suppressing background noise interference. CIoU (Complete intersection over union) is used as the bounding box regression loss function to improve the convergence speed and regression accuracy of small target localization. To adapt to the limited computing power and memory resources of the UAV's onboard computing unit, channel pruning is performed on the initial training model to remove redundant and minimally weighted network connections. Then, post-training quantization (PTQ) is used to convert the floating-point (FP32) parameters in the model to integer (INT8) parameters, generating the final lightweight horseshoe crab juvenile recognition model. Finally, the model is compiled into the TensorRT format adapted for the onboard computing unit and loaded.

[0036] According to an embodiment of the present invention, step S3 specifically includes: Once the mapping UAV has completed its flight mission, the cloud server calls upon the integrated geographic information system module to perform kernel density analysis and spatial hotspot analysis on the obtained distribution points of Chinese horseshoe crab larvae based on precise geographic coordinates, in order to quantitatively assess the spatial distribution trend of Chinese horseshoe crab larvae. Based on the spatial distribution trend of horseshoe crab larvae in China, the cloud server automatically typeset and generates a survey report, which includes three types of visualization maps: a distribution map of horseshoe crab larvae in China, a nuclear density map of horseshoe crab larvae in China, and a distribution heat map of horseshoe crab larvae in China.

[0037] In the results of the kernel density analysis, the spatial region with a kernel density value greater than or equal to 3 ind / 100 m² is extracted. The spatial region is defined as the precise survey range of Chinese horseshoe crab larvae. The edge contour of the range is vectorized and the boundary coordinate data of the precise survey range is output.

[0038] Here, the identification data includes precise geographic coordinates and close-up photos. The cloud server stores the received precise geographic coordinates and close-up photos, and supports multiple users and multiple regions to view the spatial location and image data of each juvenile horseshoe crab in real time, so as to realize the instant sharing of survey data and cross-regional collaborative analysis.

[0039] According to an embodiment of the present invention, step S4 specifically includes: The survey drone includes a built-in RTK positioning module and a full-frame mapping camera with a resolution of 60 megapixels or more and a focal length of 40 mm or more. The boundary coordinates of the precise survey area are imported into the UAV flight path planning software to generate the photogrammetric survey flight path and parameters. The specific flight path and parameters include: the flight altitude relative to the take-off point is set to 8 m, the flight speed is set to 1 m / s, the forward overlap rate is set to 60%, the lateral overlap rate is set to 50%, and the camera pitch angle is -90°. The survey drone is controlled to perform the survey mission according to the flight path. During the survey mission, the drone takes aerial photos at 1-second intervals. At each camera exposure moment, the timestamp of the exposure moment is obtained through the built-in RTK positioning module and the camera's exposure time synchronization mechanism, and the RTK antenna phase center coordinates of the UAV system at each timestamp are recorded. Based on the preset spatial offset between the RTK antenna phase center and the camera optical center, the eccentricity compensation calculation is performed on the coordinates of the RTK antenna phase center to obtain the high-precision geographic coordinates of the camera center point at the time of the corresponding aerial photo, and then written into the Exif information of the aerial photo. After the drone flight path is completed, all aerial photographs are imported into photogrammetric data analysis software. The software first extracts image feature points from overlapping areas of adjacent aerial photographs and performs corresponding point matching. Using the high-precision geographic coordinates in the Exif information of the aerial photographs as initial position constraints, aerial triangulation is performed to obtain the precise exterior orientation elements of each aerial photograph. Based on the calculated exterior orientation elements and matching feature points, a dense point cloud is generated, and a digital surface model is constructed. The digital surface model is then used to orthorectify and mosaic the aerial photographs to generate an orthophoto map covering the precise survey area.

[0040] What can be understood here is that the flight path is the same as the survey path.

[0041] According to an embodiment of the present invention, step S5 includes: The orthophoto image is input into a pre-trained segmentation model for juvenile Chinese horseshoe crabs; The orthophoto image is analyzed using an instance segmentation model to identify juvenile Chinese horseshoe crabs in the orthophoto image and extract the contours of each juvenile Chinese horseshoe crab; the pixel coordinates of each inflection point of the contour are calculated. Read the spatial reference information of the orthophoto image, wherein the spatial reference information includes at least the origin geographic coordinates and the ground sampling distance parameter; Based on the spatial reference information, an affine transformation matrix is ​​constructed from the pixel coordinate system to the real geographic coordinate system; The pixel coordinates of the contour inflection points are input into the affine transformation matrix for coordinate mapping calculation to obtain the geographic coordinates of the contour inflection points of the Chinese horseshoe crab larvae. The geographic coordinates were converted into the outline information of the Chinese horseshoe crab larvae and saved as a vector file in shp format.

[0042] According to an embodiment of the present invention, in step S5, the instance segmentation model is specifically as follows: Import multi-source orthophoto maps from the Chinese horseshoe crab larvae image dataset into a geographic information system; Annotate multi-source orthophoto maps under high-magnification display of a geographic information system, draw vector polygons along the edge of juvenile horseshoe crabs, and generate ground value labels; According to the preset slice size, the multi-source orthophoto image and the corresponding vector polygon are subjected to grid cropping and rasterization to generate an instance segmentation training set containing multiple image slices and corresponding pixel-level mask labels. An initial instance segmentation network based on the Mask R-CNN architecture is constructed. The initial instance segmentation network includes a feature extraction backbone network, a region candidate network, and a mask prediction branch. The instance segmentation training set is input into the initial instance segmentation network for iterative training. The network parameters are updated by optimizing the multi-task loss function until the network converges, thus obtaining the pre-trained Chinese horseshoe crab juvenile instance segmentation model.

[0043] According to an embodiment of the present invention, step S6 specifically includes: Importing the vector file of the juvenile Chinese horseshoe crab into a geographic information system for quantitative analysis, the quantitative analysis including at least: The number of horseshoe crab larvae in the survey area was counted, and the population density of horseshoe crab larvae was calculated based on the area parameter of the survey area. Extract the polygonal geometric features of the Chinese horseshoe crab outline, calculate the cephalothorax width of each Chinese horseshoe crab larva based on a preset measurement baseline, and infer the age of the Chinese horseshoe crab larvae based on the cephalothorax width; Spatial overlay analysis was performed on the vector profile file of Chinese horseshoe crab larvae and the distribution map of environmental factors in the survey area. Spatial clustering algorithm was used to extract the aggregation characteristics of the spatial distribution of the Chinese horseshoe crab larvae, and a spatial distribution correlation pattern between the aggregation characteristics and environmental factors such as substrate type and tidal channel distribution was established.

[0044] It can be understood here that spatial clustering algorithms may include K-Means, K-Medoids, DBSCAN, OPTICS, etc.

[0045] The following technical effects can be achieved through the embodiments of the present invention: This significantly improves survey efficiency and spatial coverage. Traditional foot surveys are limited by tidal time, resulting in short working windows and limited survey areas, making it difficult to conduct continuous surveys on a large spatial scale. UAVs can complete image acquisition and mapping of large areas of mudflats within a short tidal window, greatly expanding the coverage of a single survey and enabling continuous and comprehensive monitoring of the entire habitat.

[0046] Human interference is avoided. Ground surveys can disturb the intertidal substrate structure, directly impacting the lives of larvae and potentially distorting the data. Unmanned aerial vehicle (UAV) remote sensing, as a non-contact survey method, avoids interference with the intertidal substrate structure and horseshoe crab larvae, allowing for a more accurate reflection of their distribution and behavioral patterns in their natural state.

[0047] This method improves the reliability of survey data and reduces human error. Traditional survey methods are highly random and have a high rate of missed detections. Ultra-high resolution imagery can depict sub-millimeter-level details, overcoming the challenges of juvenile horseshoe crabs being tiny and closely resembling the color of mudflats, thus significantly reducing the missed detection rate. Combined with intelligent image recognition technology, it enables efficient and accurate analysis of massive amounts of high-definition image data, automatically identifying and counting the number of juveniles, reducing human error, and significantly improving data quality and reliability.

[0048] This ensures the personal safety of investigators. Ground-based investigators face safety risks such as tides, mudflats, and attacks from toxic or ferocious marine life. Drone surveys effectively avoid these safety risks, ensuring personal safety.

[0049] This provides technical support for the precise conservation and scientific management of horseshoe crabs in China. Through rapid surveys covering the entire region and precise regional surveys, key data such as the number of juvenile horseshoe crabs, population density, age estimation, and correlation patterns with environmental factors can be efficiently obtained, providing solid technical support for the precise conservation and scientific management of horseshoe crab juveniles in China.

[0050] This invention integrates several cutting-edge technologies, including ultra-high resolution UAV photogrammetry, remote sensing, edge computing, a lightweight YOLO target detection model, an instance segmentation model, and GIS spatial analysis, to construct a complete survey system for juvenile horseshoe crabs in China. The real-time inference detection and zoom camera close-up shooting by the UAV's onboard high-performance computing unit, the synchronous measurement by the laser rangefinder and RTK positioning module, and the real-time data sharing and visualization on the cloud server all demonstrate advanced technology integration and application capabilities.

[0051] In summary, compared with existing traditional hiking survey methods, this invention has advantages in efficiency, accuracy, environmental friendliness, personal safety, and data depth, providing an efficient and accurate technical solution for the conservation of the endangered horseshoe crab species in China.

[0052] According to an embodiment of the present invention, it further includes: For the same target object, the contour inflection points of Chinese horseshoe crab larvae are acquired in real time from multiple survey drones, and grayscale-based recognition image data is generated based on the contour inflection points. The image data of the recognition image is segmented into multiple image regions. For each image region, the Sobel operator is introduced to extract features from the recognition image data to obtain the operator feature set. Based on multiple survey drones, multiple operator feature sets were obtained; The similarity between one operator feature set and the other operator feature sets is calculated, and the recognition reliability of the survey drone is evaluated based on the similarity. Based on the reliability, the contour features of different target objects are selected from multiple survey drones to conduct statistical surveys of Chinese horseshoe crab larvae.

[0053] It should be noted that for surveys of juvenile horseshoe crabs in the same area, multiple survey drones can generally be used for simultaneous data collection. Due to the complex environment, the reliability of identified features varies. Here, we introduce Sobel operator features for the contour points of the identified object to evaluate the similarity between Sobel operator features from different drone images. High similarity indicates high reliability of the identified features obtained by that drone. For different target objects, we analyze multiple drones corresponding to the data collection time, perform reliability analysis and image feature filtering, and upload the results to a cloud server.

[0054] The similarity calculation between one operator feature set and the other operator feature sets is as follows: Select a current operator feature set and a comparison feature set, and calculate the difference in operator features for each image region between the current and comparison feature sets. The difference is calculated by converting the operator features into feature vector form for comparison, using Euclidean distance to obtain the difference for each region (the difference reflects feature similarity). The average difference is calculated to reflect the similarity between the current operator feature set and a comparison feature set. The average similarity between the current operator feature set and the other feature sets is analyzed and used as the representation of the similarity between the current operator feature set and the other operator feature sets. Multiple image region segmentation typically involves dividing an image into 3×3 regions in a grid format.

[0055] The image data for recognition is a grayscale image, which includes pixel information of the target object's outline. The outline is represented by a larger pixel value (e.g., 255), while the background is represented by a smaller pixel value (e.g., 0). Sobel feature extraction is then performed based on the grayscale image.

[0056] Figure 2 A block diagram of a Chinese horseshoe crab juvenile survey system based on remote sensing technology and intelligent image recognition is shown.

[0057] The second aspect of the present invention also provides a survey system for juvenile horseshoe crabs in China based on remote sensing technology and intelligent image recognition. The system includes: a mapping quadcopter UAV platform, a survey UAV platform, a cloud server, and a communication module. When the system is running, it implements all steps S1-S6 of the above-described survey method for juvenile horseshoe crabs in China based on remote sensing technology and intelligent image recognition.

[0058] The mapping quadcopter drone platform includes multiple mapping drones, and the survey drone platform includes multiple survey drones. A cloud server establishes a real-time communication connection with the drone platform.

[0059] In the several embodiments provided in this application, it should be understood that the disclosed devices and methods can be implemented in other ways. The device embodiments described above are merely illustrative. For example, the division of units is only a logical functional division, and in actual implementation, there may be other division methods, such as: multiple units or components can be combined, or integrated into another system, or some features can be ignored or not executed. In addition, the coupling, direct coupling, or communication connection between the various components shown or discussed can be indirect coupling or communication connection through some interfaces, devices, or units, and can be electrical, mechanical, or other forms.

[0060] The units described above as separate components may or may not be physically separate; the components shown as units may or may not be physical units. They may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0061] Furthermore, in the various embodiments of the present invention, all functional units can be integrated into one processing unit, or each unit can be a separate unit, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or in a combination of hardware and software functional units.

[0062] Those skilled in the art will understand that all or part of the steps of the above method embodiments can be implemented by hardware associated with program instructions. The aforementioned program can be stored in a computer-readable storage medium, and when executed, it performs the steps of the above method embodiments. The aforementioned storage medium includes various media capable of storing program code, such as mobile storage devices, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0063] Alternatively, if the integrated units of this invention are implemented as software functional modules and sold or used as independent products, they can also be stored in a computer-readable storage medium. Based on this understanding, the technical solutions of the embodiments of this invention, or the parts that contribute to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium and contains several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the methods described in the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as mobile storage devices, ROM, RAM, magnetic disks, or optical disks.

[0064] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for surveying juvenile horseshoe crabs in China based on remote sensing technology and intelligent image recognition, characterized in that, include: S1: Using a mapping drone, navigate and monitor within a preset area based on a preset path, and collect image data and location information in real time; S2: Target detection is performed on real-time acquired image data using a target detection algorithm. When a juvenile Chinese horseshoe crab is identified, the mapping drone performs a close-up operation on the target. Using the RTK positioning module and the synchronously acquired multi-dimensional location information, a spatial coordinate transformation algorithm is introduced to calculate the accurate geographic coordinates and upload the identification data to the cloud server. S3: Based on the identification data, conduct spatial distribution trend analysis of Chinese horseshoe crab larvae, and select the survey range and corresponding boundary coordinate data based on the density distribution; S4: Set the flight path and parameters of the survey drone according to the survey scope. During the survey, take aerial photos at preset time intervals. At the same time, based on the built-in RTK positioning module and the camera's exposure time synchronization mechanism, calculate the high-precision geographic coordinates of the center point of each aerial photo, embed the coordinates into the photo, and solve the captured aerial photos to generate orthophoto maps. S5: Import the orthophoto map into the instance segmentation model to identify Chinese horseshoe crab larvae and extract pixel-level contour features. Combine the spatial information of the orthophoto map to construct an affine transformation matrix, calculate the geographic coordinates of the inflection points in the pixel-level contour features, and form a coordinate vector file. S6: Import the coordinate vector file into the geographic information system to assess the population distribution density, age, and spatial aggregation characteristics of Chinese horseshoe crab larvae.

2. The method for surveying Chinese horseshoe crab larvae based on remote sensing technology and intelligent image recognition according to claim 1, characterized in that, Specifically, S1 is: The mapping UAV includes a visual acquisition module, a spatial positioning and ranging module, a high-speed wireless communication module, a remote sensing spectroscopic device, and a built-in high-performance computing unit; The visual acquisition module includes a visible light fixed-focus camera and a zoom camera. The visible light fixed-focus camera is used to continuously acquire video of the mudflat surface during the drone's flight, and the zoom camera is used to take close-up photos of the identified horseshoe crab larvae. The built-in high-performance computing unit establishes a communication connection with the visual acquisition module. It is equipped with a juvenile horseshoe crab intelligent recognition model and is configured to receive image data from the visible light fixed-focus camera in real time and perform target inference and detection. When a target is detected, it controls the zoom camera to move.

3. The method for surveying Chinese horseshoe crab larvae based on remote sensing technology and intelligent image recognition according to claim 2, characterized in that, S1 further includes: The laser rangefinder has a ranging accuracy of ±1 cm and a working range of 3 m to 3000 m; the visible light fixed-focus camera has a focal length of 24 mm and an effective pixel count of 20 million; the zoom camera has a variable focal length range of 30 mm to 800 mm and an effective pixel count of 20 million.

4. The method for surveying Chinese horseshoe crab larvae based on remote sensing technology and intelligent image recognition according to claim 3, characterized in that, S1 further includes: Based on the preset area, the bow-shaped flight path and preset flight parameters are set for navigation and monitoring. Specifically, the flight altitude relative to the takeoff point is set to 6 m, the flight speed is set to 5 m / s, the heading overlap rate is set to 60%, and the lateral overlap rate is set to 50%. During navigation and monitoring, the visual acquisition module of the mapping UAV continuously acquires surface image streams of the habitat mudflats at a preset frame rate; Using a built-in high-performance computing unit, the system synchronously receives surface image streams and runs a pre-loaded lightweight Chinese horseshoe crab larvae recognition model in real time to perform frame-by-frame inference detection on the image stream.

5. A method for surveying juvenile horseshoe crabs in China based on remote sensing technology and intelligent image recognition according to claim 4, characterized in that, Specifically, S2 is: The model identifies and detects juvenile horseshoe crabs in real time. If the confidence level is greater than or equal to 60%, a remote sensing spectral device is used to obtain spectral images for secondary confirmation of the object. The zoom camera is then triggered to automatically zoom. The relative position of the target in the field of view is calculated by using the two-dimensional bounding box coordinates of the target object output by the model. The gimbal is then driven to adjust its attitude and the zoom camera is controlled to automatically zoom, so that the juvenile horseshoe crab is centered and reaches the preset magnification ratio. Subsequently, the zoom camera is controlled to take a close-up image of the target object. While the zoom camera is taking the close-up image, the computing unit calls the RTK positioning module to obtain the reference geographic coordinates of the current observation device and simultaneously triggers the laser rangefinder to obtain the straight-line distance and spatial pitch and azimuth angles from the observation device to the juvenile horseshoe crab target. Based on the reference geographic coordinates, straight-line distance, spatial pitch angle and azimuth angle, the precise geographic coordinates of the Chinese horseshoe crab juvenile target are calculated by a spatial coordinate transformation algorithm, and the precise geographic coordinates are converted into metadata and embedded in the Exif information layer of the close-up photo of the target object; The close-up photo and its corresponding precise geographic coordinates are uploaded to a remote cloud server via an onboard high-speed wireless communication module.

6. The method for surveying Chinese horseshoe crab larvae based on remote sensing technology and intelligent image recognition according to claim 5, characterized in that, In S2, the spatial coordinate transformation algorithm specifically includes: A local coordinate system is established with the antenna phase center of the RTK positioning module as the origin. Combined with the attitude data obtained by the IMU inertial measurement unit built into the mapping UAV, the polar coordinate system data measured by the laser rangefinder is converted into relative spatial rectangular coordinates. Through the CGCS2000 coordinate system transformation model, the relative spatial rectangular coordinates are superimposed on the reference geographic coordinates to obtain the precise geographic coordinates of the Chinese horseshoe crab juvenile containing longitude, latitude and elevation information.

7. The method for surveying Chinese horseshoe crab larvae based on remote sensing technology and intelligent image recognition according to claim 6, characterized in that, The recognition model specifically includes: Images of juvenile horseshoe crabs under different lighting conditions and in different mudflat environments were collected. Considering the small size of the juvenile horseshoe crabs and the fact that they are generally covered with mud film, Mosaic stitching, random cropping, and adaptive histogram equalization were introduced to enhance the images. Gaussian noise was added, the image dataset was labeled, and training and testing sets were generated. Based on the YOLOv8 algorithm, the backbone feature extraction network of the original framework was replaced with the lightweight network MobileNetV3, and the attention mechanism module CBAM was introduced into the neck feature fusion layer of the network to build a recognition model. The training set is input into the recognition model for iterative training. The learning rate, batch size and number of iterations are initialized, and CIoU is used as the bounding box regression loss function. During the training process, the learning rate is dynamically adjusted according to the loss value of the validation set until the model loss function converges to obtain the initial recognition model. The initial recognition model is pruned to remove redundant and minimally weighted network connections. Post-training quantization is used to convert floating-point parameters in the model into integer parameters, generating the final lightweight Chinese horseshoe crab juvenile recognition model. The recognition model is then compiled into TensorRT format adapted for airborne computing units and loaded.

8. A method for surveying juvenile horseshoe crabs in China based on remote sensing technology and intelligent image recognition according to claim 7, characterized in that, Specifically, S3 is: Once the mapping UAV has completed its flight mission, the cloud server calls upon the integrated geographic information system module to perform kernel density analysis and spatial hotspot analysis on the obtained distribution points of Chinese horseshoe crab larvae based on precise geographic coordinates, in order to quantitatively assess the spatial distribution trend of Chinese horseshoe crab larvae. Based on the spatial distribution trend of horseshoe crab larvae in China, the cloud server automatically typeset and generates a survey report, which includes three types of visualization maps: a distribution map of horseshoe crab larvae in China, a nuclear density map of horseshoe crab larvae in China, and a distribution heat map of horseshoe crab larvae in China. In the results of the kernel density analysis, the spatial region with a kernel density value greater than or equal to 3 ind / 100 m² is extracted. The spatial region is defined as the precise survey range of Chinese horseshoe crab larvae. The edge contour of the range is vectorized and the boundary coordinate data of the precise survey range is output.

9. A method for surveying juvenile horseshoe crabs in China based on remote sensing technology and intelligent image recognition, as described in claim 8, is characterized in that... Specifically, S4 is: The survey drone includes a built-in RTK positioning module and a full-frame mapping camera with a resolution of 60 megapixels or more and a focal length of 40 mm or more. The boundary coordinates of the precise survey range are imported into the UAV flight path planning software to generate the photogrammetric survey flight path and parameters. The specific flight path and parameters include: the flight altitude relative to the take-off point is set to 8 m, the flight speed is set to 1 m / s, the forward overlap rate is set to 60%, the lateral overlap rate is set to 50%, and the camera pitch angle is -90°. The survey drone is controlled to perform survey tasks according to the flight path. During the survey tasks, the survey drone takes aerial photos at a 1-second interval. At each camera exposure moment, the timestamp of the exposure moment is obtained through the built-in RTK positioning module and the camera's exposure time synchronization mechanism, and the RTK antenna phase center coordinates of the UAV system at each timestamp are recorded. Based on the preset spatial offset between the RTK antenna phase center and the camera optical center, the eccentricity compensation calculation is performed on the coordinates of the RTK antenna phase center to obtain the high-precision geographic coordinates of the camera center point at the time of the corresponding orthophoto, and then written into the Exif information of the orthophoto. After the drone flight path is completed, all aerial photographs are imported into photogrammetric data analysis software. The software first extracts image feature points from overlapping areas of adjacent aerial photographs and performs corresponding point matching. Using the high-precision geographic coordinates in the Exif information of the aerial photographs as initial position constraints, aerial triangulation is performed to obtain the precise exterior orientation elements of each aerial photograph. Based on the calculated exterior orientation elements and matching feature points, a dense point cloud is generated, and a digital surface model is constructed. The digital surface model is then used to orthorectify and mosaic the aerial photographs to generate an orthophoto map covering the precise survey area.

10. A method for surveying juvenile horseshoe crabs in China based on remote sensing technology and intelligent image recognition according to claim 9, characterized in that, The S5 includes: The orthophoto image is input into a pre-trained segmentation model for juvenile Chinese horseshoe crabs; The orthophoto image is analyzed using an instance segmentation model to identify juvenile Chinese horseshoe crabs in the orthophoto image and extract the contours of each juvenile Chinese horseshoe crab; the pixel coordinates of each inflection point of the contour are calculated. Read the spatial reference information of the digital orthophoto image, wherein the spatial reference information includes at least the origin geographic coordinates and the ground sampling distance parameter; Based on the spatial reference information, an affine transformation matrix is ​​constructed from the pixel coordinate system to the real geographic coordinate system; The pixel coordinates of the contour inflection points are input into the affine transformation matrix for coordinate mapping calculation to obtain the geographic coordinates of the contour inflection points of the Chinese horseshoe crab larvae. The geographic coordinates were converted into the outline information of the Chinese horseshoe crab larvae and saved as a vector file in shp format.

11. A method for surveying juvenile horseshoe crabs in China based on remote sensing technology and intelligent image recognition, as described in claim 10, is characterized in that... In S5, the instance segmentation model is specifically as follows: Import multi-source orthophoto maps from the Chinese horseshoe crab larvae image dataset into a geographic information system; Annotate multi-source orthophoto maps under high-magnification display of a geographic information system, draw vector polygons along the edge of juvenile horseshoe crabs, and generate ground value labels; According to the preset slice size, the multi-source orthophoto image and the corresponding vector polygon are subjected to grid cropping and rasterization to generate an instance segmentation training set containing multiple image slices and corresponding pixel-level mask labels. An initial instance segmentation network based on the Mask R-CNN architecture is constructed. The initial instance segmentation network includes a feature extraction backbone network, a region candidate network, and a mask prediction branch. The instance segmentation training set is input into the initial instance segmentation network for iterative training. The network parameters are updated by optimizing the multi-task loss function until the network converges, thus obtaining the pre-trained Chinese horseshoe crab juvenile instance segmentation model.

12. A method for surveying juvenile horseshoe crabs in China based on remote sensing technology and intelligent image recognition, as described in claim 11, is characterized in that... Specifically, S6 is: Importing the vector file of the juvenile Chinese horseshoe crab into a geographic information system for quantitative analysis, the quantitative analysis including at least: The number of horseshoe crab larvae in the survey area was counted, and the population density of horseshoe crab larvae was calculated based on the area parameter of the survey area. Extract the polygonal geometric features of the Chinese horseshoe crab outline, calculate the cephalothorax width of each Chinese horseshoe crab larva based on a preset measurement baseline, and infer the age of the Chinese horseshoe crab larvae based on the cephalothorax width; Spatial overlay analysis was performed on the vector profile file of Chinese horseshoe crab larvae and the distribution map of environmental factors in the survey area. Spatial clustering algorithm was used to extract the aggregation characteristics of the spatial distribution of the Chinese horseshoe crab larvae, and a spatial distribution correlation pattern between the aggregation characteristics and environmental factors such as substrate type and tidal channel distribution was established.

13. A survey system for juvenile horseshoe crabs in China based on remote sensing technology and intelligent image recognition, characterized in that: The system includes: a mapping quadcopter drone platform, a survey drone platform, a cloud server, and a communication module. When the system is running, it implements the steps of the Chinese horseshoe crab juvenile survey method based on remote sensing technology and intelligent image recognition as described in claim 1.

Images