Method for estimating area of coral top
Through the band-like sample band method, a detection segmentation model was established, and a camera motion simplified model was constructed, which solved the problems of low efficiency, high cost and destructiveness of coral top area measurement, and achieved high-precision and automated coral area estimation.
Patent Information
- Application Number
- CN202510575977.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-05-06
- Publication Date
- 2025-08-15
AI Technical Summary
In the prior art, the coral top area measurement method has problems such as complex underwater environment, resulting in limited measurement, vulnerability to coral damage, low measurement efficiency, high cost and high professional skills requirements.
The band-like sample band method is used to collect the video stream data and ruler data of corals, establish a detection segmentation model, build a simplified camera motion model, calculate the actual distance and pixel distance ratio between the scale feature points, and estimate the coral top area.
A non-destructive, automated coral top area measurement is achieved, reducing labor costs and the destructive impact of surveyors on corals, and improving the accuracy and reliability of measurements.
Smart Images
Figure CN120495383A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of marine ecosystem monitoring, and in particular to a method for estimating the top area of corals. Background Art
[0002] In marine ecological research, the health of coral reefs is one of the most important research indicators, and measuring the top area of individual corals is of great significance for assessing their growth and ecological value. Traditional methods for measuring coral top area rely primarily on underwater field measurements by divers. Commonly used methods include direct measurement (divers use measuring tools to directly measure the length, width, and other dimensions of the coral top, and then calculate the top area) and photogrammetry (surveyors take photos of corals and then use image processing software to measure the coral dimensions in the photos). In addition, there are also some measurement methods based on acoustic technology, such as using sonar equipment to scan coral reefs and then analyzing and processing the acoustic information to convert it into coral size data.
[0003] However, the direct measurement method in the existing technology has shortcomings such as being limited by the complex underwater environment and the changeable topography of coral reefs, easily damaging corals, low measurement efficiency, and high professional requirements for surveyors. The photogrammetry method in the existing technology has shortcomings such as difficult camera calibration, high cost, complex data processing, and high professional skills requirements for surveyors and calculators.
[0004] Therefore, there is a need for a coral area estimation method with high precision without destroying the coral reef structure. Summary of the Invention
[0005] The main purpose of the present invention is to provide a method for estimating the coral top area, aiming to solve the problem that the existing coral area detection methods have poor measurement efficiency and measurement cost.
[0006] To achieve the above object, the present invention proposes a method for estimating the top area of coral, which comprises:
[0007] Collect coral top area data, use the strip transect method to collect coral video stream data and scale data, and build a coral dataset;
[0008] Establish a detection and segmentation model and train it on the coral dataset so that the detection and segmentation model can detect and segment corals and sample strips;
[0009] Construct a simplified camera motion model. When the geometric center of the coral is detected to be located on the midline of the image's y-axis, record the current scale feature point set and the coordinates of the coral vertices. After multiple recordings, the collected data is used to construct a simplified camera motion model.
[0010] The coral top area is estimated by calculating the ratio of the actual distance to the pixel distance between two interval ruler feature points and the camera translation distance when the ruler is at the first height, and then calculating the estimated value of the coral top area.
[0011] Preferably, the step of collecting coral top area data, collecting coral video stream data and scale data using a strip transect method, and establishing a coral dataset includes:
[0012] Using a waterproof camera, underwater light source and transect ruler, continuous photography was performed along a preset path underwater using the strip transect method;
[0013] Select an image annotation tool with a polygonal mask annotation function to mark the outlines of the coral and ruler, obtain the coral video stream data and ruler data, and construct the coral video stream data and ruler data into a coral dataset.
[0014] Preferably, the step of establishing a detection and segmentation model and training the detection and segmentation model on a coral dataset so that the detection and segmentation model can detect and segment corals and transects comprises:
[0015] Select a deep learning model for coral image segmentation and convert the coral dataset into the data format corresponding to the model;
[0016] The deep learning model was trained using the coral dataset, which was divided into corals and sample strips. When the MAP50 index of the model training reached more than 80%, the training was considered successful.
[0017] Preferably, the step of constructing a simplified camera motion model includes recording a current scale feature point set and the coordinates of the coral vertices when the geometric center of the coral is detected to be located on the midline of the y-axis of the image, and using the collected data to construct the simplified camera motion model after multiple recordings, including:
[0018] Construct a coordinate system along the camera's movement direction. When the geometric center of the coral is detected to be on the midline of the y-axis of the image, record the positions of at least two adjacent rulers in the current image, the specific spatial position of the ruler, the corresponding position of the specific spatial position of the ruler on the image, and the corresponding position of the geometric center of the coral on the image.
[0019] After the camera moves for several frames, the positions of at least two adjacent rulers after the movement, the ruler space points, the corresponding points of the ruler specific space points on the image, and the corresponding points of the coral geometric center on the image are recorded again.
[0020] Preferably, the step of estimating the coral top area by calculating the ratio of the actual distance to the pixel distance between two spaced feature points of the scale when the scale is at a first height and the translation distance of the camera to calculate the estimated value of the coral top area includes:
[0021] When the ruler is at a first height, obtaining the actual distance between any two adjacent rulers and the pixel distance between the two adjacent rulers in the image, and calculating the ratio between the actual distance and the pixel distance;
[0022] After the camera is translated along the preset path, the actual distance between any two adjacent rulers and the pixel distance between the two adjacent rulers in the image are obtained again, and the actual translation distance of the camera and the pixel translation distance between the two adjacent rulers in the image are obtained.
[0023] Preferably, the detection and segmentation model adopts the YOLOv11 architecture.
[0024] Preferably, the resolution of the coral dataset is greater than or equal to 1920*1080 pixels, the length of the ruler is greater than or equal to 3m, and texture feature points are set on the surface of the ruler.
[0025] The present invention provides a coral top area estimation method. The method collects video stream data of corals through the strip transect method and establishes a detection and segmentation model to realize automatic recognition and segmentation of corals. The method also realizes non-destructive area estimation calculation of corals through a simplified camera motion model. This reduces the destructive measurement of coral morphology by surveyors and the labor cost required for measuring coral area, which helps to better understand and predict changes in coral reef ecosystems. BRIEF DESCRIPTION OF THE DRAWINGS
[0026] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the processes shown in these drawings without paying any creative work.
[0027] Figure 1 Schematic diagram of a simplified camera motion model in one embodiment of the coral top area estimation method;
[0028] Figure 2 FIG. 1 is a flow chart of an embodiment of the coral top area estimation method. FIG. DETAILED DESCRIPTION
[0029] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.
[0030] It should be noted that all directional indications in the embodiments of the present invention (such as up, down, left, right, front, back, etc.) are only used to explain the relative position relationship, movement status, etc. between the various components under a certain specific posture (as shown in the accompanying drawings). If the specific posture changes, the directional indication will also change accordingly.
[0031] In addition, the descriptions of "first", "second", etc. in the present invention are for descriptive purposes only and should not be understood as indicating or implying their relative importance or implicitly indicating the number of the indicated technical features. Therefore, the features defined as "first" or "second" may explicitly or implicitly include at least one of such features. In addition, the technical solutions between the various embodiments can be combined with each other, but this must be based on the fact that they can be implemented by ordinary technicians in this field. When the combination of technical solutions is contradictory or cannot be implemented, it should be deemed that such combination of technical solutions does not exist and is not within the scope of protection required by the present invention.
[0032] In the existing technology, when it comes to measuring coral area, traditional measurement methods have great limitations. Direct measurement methods are prone to measurement restrictions in complex underwater environments. The complex underwater environment and the changeable terrain of coral reefs make it difficult for surveyors to conduct direct measurements. When surveyors use measuring tools to measure corals, they are also likely to cause damage to the corals. Because they are in an underwater environment, multiple divers are required to conduct measurements, which places great demands on the number and professional capabilities of divers and increases labor costs. Photogrammetry requires precise calibration of corals through cameras, but in underwater environments, the camera calibration process is very complicated and requires the use of specific calibration plates and complex algorithms. It also places extremely high demands on the professional capabilities of surveyors, resulting in extremely high measurement costs, poor measurement accuracy, and insufficient robustness of the measurement method.
[0033] Based on this, refer to Figure 1-Figure 2 The present invention provides a method for estimating the top area of coral, which specifically includes the following steps:
[0034] S10, collecting coral top area data, using the strip transect method to collect coral video stream data and scale data, and establishing a coral dataset;
[0035] S20, establishing a detection and segmentation model, and training the detection and segmentation model on the coral dataset so that the detection and segmentation model can detect and segment corals and sample strips;
[0036] S30, constructing a simplified camera motion model. When the geometric center of the coral is detected to be located on the midline of the image y-axis, the current scale feature point set and the coordinates of the coral vertices are recorded. After multiple recordings, the collected data is used to construct the simplified camera motion model.
[0037] S40, estimating and measuring the coral top area, calculating the ratio of the actual distance to the pixel distance between two spaced feature points of the ruler when the ruler is at a first height and the translation distance of the camera, and calculating an estimated value of the coral top area.
[0038] In this embodiment, the coral video stream data collected must include images of different coral types, different lighting conditions, and background environments. The images must contain clear corals and rulers. Furthermore, the dataset must contain at least 5,000 images (i.e., the video stream data must contain at least 5,000 frames of video) to ensure that the subsequent training model can meet generalization requirements. It is understood that if an existing coral image dataset is available, it can be used directly.
[0039] Specifically, a waterproof camera needs to be selected and adapted to the corresponding underwater light source to ensure that the waterproof camera can obtain clear images. At the same time, the sample strip ruler also needs to have clear scales, and obvious marks need to be set at evenly spaced intervals on the sample strip ruler to facilitate model learning.
[0040] The waterproof camera uses a strip transect method to capture video underwater, continuously filming along a pre-set transect path. The transect path should cover different areas of the coral reef to ensure representative image data is collected. The waterproof camera should be kept between 1 and 3 meters from the coral.
[0041] In one embodiment, the step of S10, collecting coral top area data, using a strip transect method to collect coral video stream data and scale data, and establishing a coral dataset, includes:
[0042] S11. Use a waterproof camera, underwater light source, and transect ruler, and take continuous photos underwater along a preset path using the strip transect method.
[0043] S12. Select an image annotation tool with a polygon mask annotation function to annotate the outlines of the coral and the ruler, obtain the video stream data and the ruler data of the coral, and construct the video stream data and the ruler data of the coral into a coral dataset.
[0044] In this example, the image annotation tools selected for polygon mask annotation are Labelme and Taglab. It is understood that other image annotation tools available to the operator include, but are not limited to, Labelme and Taglab. During the annotation process, the edge contours of the coral and the transect scale should be marked. The final annotation file should contain the coordinates of the marked polygon and the classification information.
[0045] In one embodiment, the step of S20, establishing a detection and segmentation model, and training the detection and segmentation model on a coral dataset so that the detection and segmentation model can detect and segment corals and transects, includes:
[0046] S21. Select a deep learning model for coral image segmentation and convert the coral dataset into a data format corresponding to the model;
[0047] S22. Use the coral dataset to train the deep learning model. Split the coral dataset into corals and sample strips. When the MAP50 index of the model training reaches more than 80%, the training is considered successful.
[0048] In this embodiment, a segmentation model based on the YOLOv11 architecture is constructed to detect ruler positioning and coral instance segmentation. When the validation set mAP@50 (mAP@50 refers to the mean average precision (mAP) evaluation indicator in the object detection task, where the IoU threshold is set to 0.5) of the model training reaches 80% or above, the training is considered successful and the training is stopped.
[0049] In another embodiment, the present invention may also set a dynamic learning rate, which is initially 3e-4 and decays by 30% every 10 epochs.
[0050] In one embodiment, S30, constructing a simplified camera motion model, includes recording a current scale feature point set and the coordinates of the coral vertices when the geometric center of the coral is detected to be located on the midline of the image y-axis, and using the collected data to construct the simplified camera motion model after multiple recordings.
[0051] S31. Construct a coordinate system along the camera movement direction. When the geometric center of the coral is detected to be on the midline of the y-axis in the image, record the points of at least two adjacent rulers in the current image, the specific spatial point of the ruler, the corresponding point of the specific spatial point of the ruler on the image, and the corresponding point of the geometric center of the coral on the image;
[0052] S32. After the camera moves for several frames, the positions of at least two adjacent rulers after the movement, the ruler spatial positions, the corresponding positions of the ruler specific spatial positions on the image, and the corresponding position of the coral geometric center on the image are recorded again.
[0053] In this embodiment, when the geometric center of the coral is detected to be on the midline of the y-axis in the image, the adjacent points 1 and 2 on the sample scale are recorded, which are specifically represented as: point 1 P ruler1 、Point 2P rul , scale space point C center , the corresponding point P of the specific spatial point of the ruler on the image Center , the corresponding point P of the coral geometric center on the image coral_center .
[0054] After several frames, the adjacent point 1 and point 2 and other data are captured and recorded again, specifically expressed as: point 1 P′ ruler1 , point 2 P′ ruler2 , scale space point C′ center , the corresponding point P′ of the specific spatial point of the ruler on the image center , the corresponding point P′ of the geometric center of the coral on the image coral_center .
[0055] In one embodiment, S40, estimating the coral top area, includes calculating the ratio of the actual distance to the pixel distance between two spaced apart feature points of the scale when the scale is at a first height, and the camera translation distance, to calculate an estimated value of the coral top area, including:
[0056] S41. When the scale is at a first height, obtain the actual distance between any two adjacent scales and the pixel distance between the two adjacent scales in the image, and calculate the ratio between the actual distance and the pixel distance;
[0057] S42. After the camera is translated along the preset path, the actual distance between any two adjacent rulers and the pixel distance between the two adjacent rulers in the image are obtained again, and the actual translation distance of the camera and the pixel translation distance between the two adjacent rulers in the image are obtained.
[0058] In this embodiment, when the scale is at the initial height H ruler Calculate the ratio between the actual distance and pixel distance between point 1 and point 2 on the transect scale. ruler_H .
[0059] The specific formula is as follows:
[0060]
[0061] ruler_image=Distance(P′ ruler1 ,P′ ruler2 )
[0062] ruler_reality=10cm
[0063] Scale ruler_H =H ruler / f
[0064] Among them, ruler reality is the actual distance between point 1 and point 2, ruler image is the speed limit distance between point 1 and point 2, Scale ruler_H =H ruler / f is the camera model, and f is the camera intrinsic focal length.
[0065] After the camera moves along the preset path (i.e., the Y direction of the transect scale), the camera's moving path space is parallel to the extension direction of the transect scale, and the camera optical axis remains perpendicular to the transect scale. By calculating the distance Y that the corresponding camera moves along the transect scale in the two frames of image, cam This is equivalent to calculating the actual translation distance between point 1 and point 2. The specific formula is as follows:
[0066] dY center_img =P center_y -P′ center_y
[0067]
[0068] Among them, P′ center_y P′ center Y coordinate value, P center_y P center The Y coordinate value, Y center_img P center and P′ center The pixel distance in the Y direction. After the camera moves, the height of the coral detected is recorded as H coral , the estimated ratio of the actual distance and the pixel distance is recorded as Scale coral_H_est , the specific formula is as follows:
[0069]
[0070]
[0071] Area_reality coral_est =Area_imager coral *Scale coral_H_est 2
[0072] Among them, dY coral_img Area_reality is the pixel movement distance of the coral center. coral_est is an estimate of the coral top area.
[0073] In one embodiment, the detection and segmentation model adopts the YOLOv11 architecture.
[0074] In detail, the YOLOv11 architecture can adopt a transfer learning strategy. By loading pre-trained models, configuring loss functions, configuring dynamic learning rates, and setting training termination conditions, it can optimize model efficiency while avoiding overfitting and improving model robustness.
[0075] In one embodiment, the resolution of the coral dataset is greater than or equal to 1920*1080 pixels, the length of the ruler is greater than or equal to 3 meters, and texture feature points are set on the surface of the ruler.
[0076] In this embodiment, the texture feature points set on the surface of the ruler have obvious marks and are set every 10 cm on the sample strip ruler.
[0077] In another embodiment, the texture feature points are set at intervals of 5 cm. It is understandable that the interval threshold of the texture feature points is 5-10 cm, ensuring that the waterproof camera can clearly and accurately capture the texture feature points during movement.
[0078] In combination with all the above embodiments, the present invention provides a method for estimating the top area of corals. Video stream data of corals is collected through a strip transect method and a detection and segmentation model is established to achieve automatic identification and segmentation of corals. A camera motion simplification model is used to achieve non-destructive area estimation and calculation of corals. Automated processing reduces the impact of human factors on data processing, improves the reliability and consistency of data processing, ensures the accuracy and repeatability of each measurement result, reduces the destructive measurement of coral morphology by surveyors and the labor cost required to measure coral area, and contributes to a better understanding and prediction of changes in coral reef ecosystems.
[0079] The above description is only a preferred embodiment of the present invention and does not limit the patent scope of the present invention. All equivalent structural transformations made by using the contents of the present invention description and drawings under the inventive concept of the present invention, or direct / indirect application in other related technical fields are included in the patent protection scope of the present invention.
Claims
1. A method for estimating the top area of coral, characterized in that: The following steps are involved: Collect coral top area data, use the strip transect method to collect coral video stream data and scale data, and build a coral dataset; Establish a detection and segmentation model and train it on the coral dataset so that the detection and segmentation model can detect and segment corals and sample strips; Construct a simplified camera motion model. When the geometric center of the coral is detected to be located on the midline of the image's y-axis, record the current scale feature point set and the coordinates of the coral vertices. After multiple recordings, the collected data is used to construct a simplified camera motion model. The coral top area is estimated by calculating the ratio of the actual distance to the pixel distance between two interval ruler feature points and the camera translation distance when the ruler is at the first height, and then calculating the estimated value of the coral top area.
2. The method for estimating the coral top area according to claim 1, wherein: The steps of collecting coral top area data, collecting coral video stream data and scale data using a strip transect method, and establishing a coral data set include: Using a waterproof camera, underwater light source and transect ruler, continuous photography was performed along a preset path underwater using the strip transect method; Select an image annotation tool with a polygonal mask annotation function to mark the outlines of the coral and ruler, obtain the coral video stream data and ruler data, and construct the coral video stream data and ruler data into a coral dataset.
3. The method for estimating the coral top area according to claim 2, wherein: The steps of establishing a detection and segmentation model and training the detection and segmentation model in a coral dataset so that the detection and segmentation model can detect and segment corals and transects include: Select a deep learning model for coral image segmentation and convert the coral dataset into the data format corresponding to the model; The deep learning model was trained using the coral dataset, which was divided into corals and sample strips. When the MAP50 index of the model training reached more than 80%, the training was considered successful.
4. The method for estimating the coral top area according to claim 1, wherein: The step of constructing a simplified camera motion model, when detecting that the geometric center of the coral is located on the midline of the y-axis of the image, recording the current scale feature point set and the coordinates of the coral vertices, and using the collected data to construct the simplified camera motion model after multiple recordings, includes: Construct a coordinate system along the camera's movement direction. When the geometric center of the coral is detected to be on the midline of the y-axis of the image, record the positions of at least two adjacent rulers in the current image, the specific spatial position of the ruler, the corresponding position of the specific spatial position of the ruler on the image, and the corresponding position of the geometric center of the coral on the image. After the camera moves for several frames, the positions of at least two adjacent rulers after the movement, the ruler space points, the corresponding points of the ruler specific space points on the image, and the corresponding points of the coral geometric center on the image are recorded again.
5. The method for estimating the coral top area according to claim 1, wherein: The coral top area estimation and calculation step includes calculating the ratio of the actual distance to the pixel distance between two spaced feature points of the ruler when the ruler is at a first height and the translation distance of the camera to calculate the estimated value of the coral top area, including: When the ruler is at a first height, obtaining the actual distance between any two adjacent rulers and the pixel distance between the two adjacent rulers in the image, and calculating the ratio between the actual distance and the pixel distance; After the camera is translated along the preset path, the actual distance between any two adjacent rulers and the pixel distance between the two adjacent rulers in the image are obtained again, and the actual translation distance of the camera and the pixel translation distance between the two adjacent rulers in the image are obtained.
6. The method for estimating the coral top area according to claim 1, wherein: The detection and segmentation model adopts the YOLOv11 architecture.
7. The method for estimating the coral top area according to claim 1, wherein: The resolution of the coral dataset is greater than or equal to 1920*1080 pixels, the length of the ruler is greater than or equal to 3m, and texture feature points are set on the surface of the ruler.