A method for constructing a deep-sea benthic organism feature database integrating environmental information

Abstract

This invention discloses a method for constructing a deep-sea benthic biota feature library that integrates environmental information. The method includes the following steps: classifying and organizing original benthic biota photographs and recording sampled environmental feature data; determining the target biota image area and the environmental background image area in each photograph; using computer image processing techniques to extract image attributes such as the biota morphology and body color, while also extracting image attributes representing the biota's habitat; and employing a clustering algorithm to analyze each type of feature dataset by biota group to construct a deep-sea benthic biota feature library containing multiple attributes. The method has the advantage of enabling a more accurate, comprehensive, and systematic description of benthic biota targets, and can be used to obtain highly reliable basic data required for subsequent benthic biota feature research and automatic identification applications.

Description

Technical Field

[0001] The present invention relates to the fields of seabed ecological environment investigation and scientific research, computer image processing, and in particular to a method for constructing a deep-sea benthic organism feature library integrating environmental information. Background Art

[0002] Benthic organisms are widespread and extensive, fine-scale targets on the deep seafloor. They are also key targets for potential deep-sea resources and environmental protection. During ocean ecological surveys, the primary method for detecting deep-sea benthic targets currently involves first capturing images of benthic targets using unmanned underwater systems such as optical camera tows, autonomous underwater vehicles (AUVs), tethered underwater vehicles (ROVs), and deep-sea landers. Biologists then perform offline manual identification and category labeling based on the collected images. In recent years, a small number of applications for automatic identification of marine biological targets based on artificial intelligence technologies such as deep learning have been developed. However, due to the relative scarcity of live footage of deep-sea benthic organisms and the generally poor quality of available online images, the number of effective benthic images is extremely limited, significantly impacting the advantages and computational performance of deep learning algorithms. Furthermore, when implementing deep learning-based automatic identification algorithms for biological targets, intermediate processing steps such as target feature extraction and description are not explicitly performed independently. Consequently, the output of benthic target feature modeling and database construction cannot be directly generated, hindering the further application of these algorithms in biological signature research. In addition, traditional methods for constructing benthic biological feature libraries often only mine the image contours and other content of biological targets for feature modeling, but ignore the sampling environment characteristics when the biological photos were taken. These environmental characteristics are crucial to the description of biological habitats. Making full use of them can further improve the accuracy of automatic identification of benthic organisms. Summary of the Invention

[0003] To address the issues of deep-sea benthic organism feature modeling and feature library construction, this paper proposes a deep-sea benthic organism feature library construction method that integrates environmental information. By performing image processing and analysis on real-life photos of deep-sea benthic organisms of different groups, individual attribute characteristics such as morphology and body color are extracted. At the same time, attribute characteristics of the benthic organisms' habitats are mined, and cluster analysis is performed on different types of feature data values. Ultimately, a deep-sea benthic organism feature database containing multiple attributes is constructed. The method specifically includes the following steps:

[0004] (1) Classify and organize the original benthic photos collected during the voyage and record the sampling environment characteristic data of each photo;

[0005] (2) According to the catalog of deep-sea benthic organisms, all biological target image areas and environmental background image areas in each original benthic organism photo were determined, and the HSV color mean and scale-invariant feature transform (SIFT) key point feature matrix corresponding to the environmental background image area were extracted and recorded as environmental background image features;

[0006] (3) Use the image interpolation algorithm to convert the biological target image into a uniform size, and extract the HSV color mean, SIFT key point feature matrix, and border outline array of each biological target in turn, which are recorded as the biological target image features;

[0007] (4) Use clustering algorithms to perform statistical analysis on the sampling environment characteristics, environmental background image characteristics, and biological target image characteristics of different biological groups, and save the statistical results of the feature data in the form of a database.

[0008] Furthermore, step (1) specifically includes the following steps:

[0009] (1-1) Manually classify and screen the original benthic organism photos according to specific biological groups, and set up separate photo directories for different groups to store all benthic organism photos belonging to that group;

[0010] (1-2) If the same photo contains benthic organisms of different taxa, the photo needs to be classified into multiple corresponding taxa at the same time;

[0011] (1-3) Create a table file to record the sampling environment characteristic data corresponding to each benthic organism photo, including the group name, photo file name, water depth value, water temperature, longitude and latitude. The water depth value, water temperature, longitude and latitude data are obtained from the temperature and salinity depth meter (CTD) and underwater positioning beacon instrument recording data during underwater operations.

[0012] Furthermore, step (2) specifically includes the following steps:

[0013] (2-1) Based on the classification results obtained in step (1-2), for the original benthic organism photo in the currently selected biological group catalog, use multiple variable-size rectangular frames to encircle all biological target image areas belonging to the current biological group catalog in the original benthic organism photo, the size of each rectangular frame being determined according to the size of the area occupied by the encircled organism, and the area outside the rectangular frame is the environmental background image area;

[0014] (2-2) Using the GrabCut algorithm, the biological target image area enclosed by the rectangular frame is segmented from the original benthic organism photo, and the image segmentation result is saved as a new biological target image file for further processing and analysis;

[0015] (2-3) For the pixel points in the environment background image area, convert them from RGB color space to HSV color space, and calculate the statistical mean BC of the HSV color components of the environment background image respectively i :

[0016]

[0017] Among them, BC i is the statistical mean of the i-th HSV color component of the environmental background image; N BC Indicates the total number of pixels in the environmental background image area, bc i，j Represents the i-th HSV color component corresponding to the j-th pixel in the environmental background image;

[0018] (2-4) For the pixel points in the environmental background image area, convert them from the RGB color space to grayscale values, extract all SIFT key points in the area, and obtain a matrix composed of their feature vectors.

[0019] Furthermore, step (3) specifically includes the following steps:

[0020] (3-1) For the biological target image file obtained after image segmentation in step (2-2), a bilinear interpolation algorithm is used to adjust the image to a uniform height and width, and the adjusted image size is M pixels × M pixels;

[0021] (3-2) For the interpolated biological target image, convert it from RGB color space to HSV color space, and calculate the statistical mean TC of the color components of the biological target image respectively. i :

[0022]

[0023] Among them, TC i is the statistical mean of the i-th HSV color component of the biological target image; N TC Indicates the total number of pixels in the biological target image, tc i，j Represents the i-th HSV color component corresponding to the j-th pixel in the biological target image;

[0024] (3-3) For the interpolated biological target image, convert it from RGB color space to grayscale value, extract all SIFT key points in the image, and obtain the matrix composed of its feature vectors;

[0025] (3-4) For the interpolated biological target image, the Canny edge detection operator is used to obtain the corresponding binary image, and the border outline array of the biological target is detected and extracted from it;

[0026] (3-5) According to step (3-1) to step (3-4), other biological target images belonging to the current biological group in the current original benthic biological photo are processed in sequence to obtain the HSV color mean, SIFT key point feature matrix, and border outline array of the corresponding biological target image;

[0027] (3-6) According to steps (2-1) to (3-5), other original benthic organism photos in the current biological group catalog are processed in sequence to obtain environmental background feature data corresponding to the original benthic organism photos and feature data of all biological target images.

[0028] Furthermore, step (4) specifically includes the following steps:

[0029] (4-1) Create a unified SQLite blank database for all biological groups, define the database table structure, including the names of biological groups and all subordinate feature names in the sampling environment characteristics, environmental background image characteristics and biological target image characteristics of benthic organisms, and define corresponding data types for all names;

[0030] (4-2) Based on the record table created in step (1-3), the sampling environment characteristics, environmental background image characteristics, and biological target image characteristics corresponding to all original benthic organism photos under the biological group are selected to form three types of feature data sets respectively;

[0031] (4-3) Use the K-means algorithm to perform cluster analysis on each feature in each feature data set, obtain the cluster center corresponding to each feature as the characteristic quantitative value describing the current biological group, and write the results into the SQLite database;

[0032] (4-4) Repeat steps (4-2) to (4-3), process the three types of characteristic data sets under each biological group directory in turn, obtain the characteristic quantitative values ​​of the corresponding biological groups, and establish a complete benthic biological optical characteristic database.

[0033] Beneficial effects of the present invention: Based on computer image processing technology, the present invention extracts and analyzes multiple attributes of biological target images and environmental background images in real photos of deep-sea benthic organisms. At the same time, it establishes a characteristic model of the biological target from multiple dimensions such as morphology, body color, and habitat. With the help of data statistical analysis methods, each type of feature is quantitatively represented and a feature database is constructed. Thanks to the mining and utilization of the characteristics of the habitats where benthic organisms are located, the method of the present invention can more accurately, comprehensively, and systematically describe benthic biological targets than existing methods, further improving the data quality of the benthic biological feature database and providing a new technical means for obtaining high-reliability basic data required for subsequent benthic biological feature research and automatic identification applications. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] Figure 1 It is a specific flow chart of the method of the present invention. DETAILED DESCRIPTION

[0035] In order to describe the present invention more specifically, the technical solution of the present invention is described in detail below with reference to the accompanying drawings and specific embodiments.

[0036] like Figure 1 As shown, an embodiment of the present invention provides a method for constructing a deep-sea benthic organism feature library integrating environmental information, comprising the following steps:

[0037] Step 1 involves manually classifying and screening the original benthic photos according to specific taxa. Separate catalogs are created for each taxon, storing all photos of benthic organisms belonging to that taxon. If a single deep-sea benthic photograph contains benthic organisms from multiple taxa, the photo must be classified into multiple taxon catalogs. For example, a benthic biota signature database for the Tsaiwei Seamount in the Pacific Ocean was constructed. A statistical survey revealed that the biota system in this seamount comprises six phyla, 38 families, and 63 species of giant benthic organisms. Therefore, the taxon catalogs were created at the family level.

[0038] Step 2: Create a table file to record the sampling environment characteristic data corresponding to each benthic organism photo, including the group name, photo file name, water depth value, water temperature, longitude and latitude. The water depth value, water temperature, longitude and latitude data are obtained from the temperature and salinity depth meter (CTD) and underwater positioning beacon instrument recording data during underwater operations.

[0039] Step 3: Based on the classification results obtained in step 1, for the original benthic organism photos in the currently selected biological group catalog, use multiple variable-size rectangular frames to circle all biological target image areas belonging to the current biological group catalog in the original benthic organism photos. The size of each rectangular frame is determined according to the size of the area occupied by the encircled organism, and the area outside the rectangular frame is the environmental background image area.

[0040] Step 4: Use the GrabCut algorithm to segment the biological target image area enclosed by the rectangular frame from the original benthic organism photo, and save the image segmentation result as a new biological target image file for further processing and analysis.

[0041] Step 5: For the pixel points in the environment background image area, convert them from the RGB color space to the HSV color space of hue, saturation and brightness, and calculate the statistical mean BC of the color components of the environment background image respectively. i :

[0042]

[0043] Among them, BC i is the statistical mean of the i-th HSV color component of the environmental background image; N BC Indicates the total number of pixels in the environmental background image area, bc i，j Represents the i-th HSV color component corresponding to the j-th pixel in the environmental background image.

[0044] Step 6: For the pixel points in the environmental background image area, convert them from RGB color space to grayscale values, extract all SIFT key points in the area, and obtain a matrix composed of their feature vectors.

[0045] Step 7: For the biological target image file obtained after image segmentation in step 4, a bilinear interpolation algorithm is used to adjust the image to a uniform height and width. Considering that the number of extracted SIFT feature points is moderate, the adjusted image size is set to 200 pixels × 200 pixels.

[0046] Step 8: For the interpolated biological target image, convert it from RGB color space to HSV color space, and calculate the statistical mean TC of the color components of the biological target image respectively. i :

[0047]

[0048] Among them, TC i is the statistical mean of the i-th HSV color component of the biological target image; N TC Indicates the total number of pixels in the biological target image, tc i，j Represents the i-th HSV color component corresponding to the j-th pixel in the biological target image.

[0049] Step 9: For the interpolated biological target image, convert it from the RGB color space to grayscale values, extract all SIFT key points in the image, and obtain a matrix composed of its feature vectors.

[0050] Step 10: For the interpolated biological target image, a Canny edge detection operator is used to obtain a corresponding binary image, and a border outline array of the biological target is detected and extracted from the binary image.

[0051] Step 11: According to steps 7 to 10, other biological target images belonging to the current biological group in the current original benthic biological photo are processed in sequence to obtain the HSV color mean, SIFT key point feature matrix, and border outline array of the corresponding biological target image.

[0052] Step 12: According to steps 5 to 11, other original benthic organism photos in the current biological group catalog are processed in sequence to obtain environmental background feature data corresponding to the original benthic organism photos and feature data of all biological target images.

[0053] Step 13, create a unified SQLite blank database for all biological groups, define the database table structure, including the biological group name and the sampling environment characteristics of benthic organisms, all subordinate feature names in the environmental background image characteristics and biological target image characteristics, and define corresponding data types for all names. The subordinate features of the sampling environment characteristics are the group name, photo file name, water depth value, water temperature and longitude and latitude. The subordinate features of the environmental background image characteristics are the color component statistical mean and SIFT key points. The subordinate features of the biological target image characteristics are the color component statistical mean, SIFT key points and the border outline array of the biological target.

[0054] Step 14, based on the record table created in step 2, select the sampling environment characteristics, environmental background image characteristics, and biological target image characteristics corresponding to all original benthic organism photos under the biological group, and form three types of feature data sets respectively.

[0055] Step 15: Use the K-means algorithm to perform cluster analysis on each feature under each major feature data set, obtain the cluster center corresponding to each feature as a general feature quantitative value describing the current biological group, and write the results into the SQLite database.

[0056] Step 16, repeating steps 14 and 15, processing the three types of characteristic data sets under each biological group directory in turn, obtaining the general characteristic quantitative values ​​of the corresponding biological groups, and establishing a complete benthic organism optical characteristic database.

[0057] The specific implementation methods described above provide a detailed description of the technical solutions and beneficial effects of the present invention. It should be understood that the above is only the most preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, supplements and equivalent substitutions made within the scope of the principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A method for constructing a deep-sea benthic organism feature database integrating environmental information, characterized in that: The following steps are involved: (1) Classify and organize the original benthic photos collected during the voyage, and set up separate photo directories for different groups to store all photos of benthic organisms belonging to that group, and record the sampling environment characteristic data of each photo; the content includes the group name, photo file name, water depth, water temperature, longitude and latitude; (2) According to the catalog of deep-sea benthic biological groups, all biological target image areas and environmental background image areas in each original benthic biological photo were determined, the biological target image areas were segmented from the original benthic biological photos, and the HSV color mean and scale-invariant feature transform (SIFT) key point feature matrix corresponding to the environmental background image area were extracted and recorded as environmental background image features; (3) Use the image interpolation algorithm to convert the biological target image into a uniform size, and extract the HSV color mean, SIFT key point feature matrix, and border outline array of each biological target in turn, which are recorded as the biological target image features; (4) Use clustering algorithms to perform statistical analysis on the sampling environment characteristics, environmental background image characteristics, and biological target image characteristics of different biological groups, and save the statistical results of the feature data in the form of a database.

2. The method for constructing a deep-sea benthic organism feature database integrating environmental information according to claim 1, wherein: Step (1) specifically includes the following steps: (1-1) Manually classify and screen the original benthic photos according to specific biological groups; (1-2) If the same photo contains benthic organisms of different taxa, the photo needs to be classified into multiple corresponding taxa at the same time; (1-3) Create a table file to record the sampling environment characteristic data corresponding to each benthic organism photo. The water depth value, water temperature, and longitude and latitude data are obtained from the temperature-salinity-depth meter (CTD) and underwater positioning beacon instrument recording data during underwater operations.

3. The method for constructing a deep-sea benthic organism feature database integrating environmental information according to claim 2, wherein: Step (2) specifically includes the following steps: (2-1) Based on the classification results obtained in step (1-2), for the original benthic organism photo in the currently selected biological group catalog, use multiple variable-size rectangular frames to encircle all biological target image areas belonging to the current biological group catalog in the original benthic organism photo, the size of each rectangular frame being determined according to the size of the area occupied by the encircled organism, and the area outside the rectangular frame is the environmental background image area; (2-2) Using the GrabCut algorithm, the biological target image area enclosed by the rectangular frame is segmented from the original benthic organism photo, and the image segmentation result is saved as a new biological target image file for further processing and analysis; (2-3) For the pixel points in the environment background image area, convert them from RGB color space to HSV color space, and calculate the statistical mean BC of the HSV color components of the environment background image respectively i : Among them, BC i is the statistical mean of the i-th HSV color component of the environmental background image; N BC Indicates the total number of pixels in the environmental background image area, bc i,j Represents the i-th HSV color component corresponding to the j-th pixel in the environmental background image; (2-4) For the pixel points in the environmental background image area, convert them from the RGB color space to grayscale values, extract all SIFT key points in the area, and obtain a matrix composed of their feature vectors.

4. The method for constructing a deep-sea benthic organism feature database integrating environmental information according to claim 3, wherein: Step (3) specifically includes the following steps: (3-1) For the biological target image file obtained after image segmentation in step (2-2), a bilinear interpolation algorithm is used to adjust the image to a uniform height and width, and the adjusted image size is M pixels × M pixels; (3-2) For the interpolated biological target image, convert it from RGB color space to HSV color space, and calculate the statistical mean TC of the color components of the biological target image respectively. i : Among them, TC i is the statistical mean of the i-th HSV color component of the biological target image; N TC Indicates the total number of pixels in the biological target image, tc i,j Represents the i-th HSV color component corresponding to the j-th pixel in the biological target image; (3-3) For the interpolated biological target image, convert it from RGB color space to grayscale value, extract all SIFT key points in the image, and obtain the matrix composed of its feature vectors; (3-4) For the interpolated biological target image, the Canny edge detection operator is used to obtain the corresponding binary image, and the border outline array of the biological target is detected and extracted from it; (3-5) According to step (3-1) to step (3-4), other biological target images belonging to the current biological group in the current original benthic biological photo are processed in sequence to obtain the HSV color mean, SIFT key point feature matrix, and border outline array of the corresponding biological target image; (3-6) According to steps (2-1) to (3-5), other original benthic organism photos in the current biological group catalog are processed in sequence to obtain environmental background feature data corresponding to the original benthic organism photos and feature data of all biological target images.

5. The method for constructing a deep-sea benthic organism feature database integrating environmental information according to claim 4, characterized in that: Step (4) specifically includes the following steps: (4-1) Create a unified SQLite blank database for all biological groups, define the database table structure, including the names of biological groups and all subordinate feature names in the sampling environment characteristics, environmental background image characteristics and biological target image characteristics of benthic organisms, and define corresponding data types for all names; (4-2) Based on the record table created in step (1-3), the sampling environment characteristics, environmental background image characteristics, and biological target image characteristics corresponding to all original benthic organism photos under the biological group are selected to form three types of feature data sets respectively; (4-3) Use the K-means algorithm to perform cluster analysis on each feature under each type of feature data set separately, obtain the cluster center corresponding to each feature as the characteristic quantitative value describing the current biological group, and write the result into the SQLite database; (4-4) Repeat steps (4-2) to (4-3), process the three types of feature data sets under each biological group directory in turn, obtain the characteristic quantitative values ​​of the corresponding biological groups, and establish a complete benthic biological optical feature database.

Images