Underwater image data set generation method and device based on physical degradation

Abstract

The invention relates to an underwater image data set generation method and device based on real physical degradation, and belongs to the field of computer vision. The device comprises a transparent water tank, an underwater light supplementing lamp, a camera, an adjustable support, a computer and the like. Different underwater degradation scenes are simulated by adjusting suspended particle concentration, light intensity and camera parameters, and underwater images of various degradation degrees are collected. A clear image is directly shot, and a degraded image is acquired by adjusting environmental conditions, so that distortion caused by algorithm generation is avoided. After data acquisition, the image is cut, aligned and paired through an algorithm, and experimental parameter labels are added. And finally storing the data according to a unified rule and format, and supporting JSON or CSV metadata recording. The underwater image data set generated by the method is suitable for underwater image enhancement, target detection and segmentation tasks, and has the advantages of strong sense of reality, controllable parameters and high experiment reproducibility.

Description

Technical Field

[0001] The present invention relates to the technical field of computer vision and image processing. Specifically, it relates to a method and device for generating an underwater image dataset based on physical degradation simulation, which is applicable to research fields such as underwater image enhancement, target detection, and image segmentation. Background Art

[0002] Currently, underwater image processing plays an important role in computer vision and is widely used in scenarios such as underwater detection, robot navigation, and ocean monitoring. However, due to the attenuation of light, color shift, and interference from suspended particles in the underwater environment, the acquired underwater images have degradation problems such as blurring and color distortion. To solve these problems, researchers have developed multiple underwater image datasets to provide a basis for the research of image enhancement algorithms. The existing main underwater image datasets (such as UIEB, OceanDark, EUVP, and UFO-120) all generate reference clear images through algorithms. Some datasets (such as EUVP and UFO-120) convert clear images into blurred images through degradation models. All of these reference images are generated based on algorithms, lacking physical authenticity, difficult to cover diverse underwater environments, and limiting the practical application effect of enhancement algorithms. Therefore, there is an urgent need for a method to generate a realistic underwater image dataset through physical degradation means to solve the deficiencies in the prior art. Summary of the Invention

[0003] The present invention aims to solve the problems that the reference images of existing underwater image datasets rely on algorithm generation, are not realistic enough, and the degradation effect does not conform to the actual underwater environment. By constructing a realistic physical degradation underwater scene, generating degraded images with high realism, and providing paired clear images at the same time, a high-quality underwater image dataset is constructed to provide more reliable basic data support for underwater image enhancement and related research.

[0004] To achieve the above object, the technical solution adopted by the present invention is as follows:

[0005] (1) Device part

[0006] The main function of this device is to simulate the degradation phenomenon in a real underwater environment and collect corresponding clear images and degraded images. The device mainly includes a pool, a light source, a camera, a fixed bracket, and a collection device. By adjusting the water turbidity, lighting conditions, and shooting parameters of the camera, the device can provide a high-quality and controllable experimental environment for the construction of an underwater image dataset. The device supports the simulation of multiple underwater degradation scenarios, such as water bodies with different turbidities, uneven lighting environments, and changes in the distance and angle of target objects. The device can not only accurately collect clear images but also generate highly realistic degraded images through real physical degradation means.

[0007] Preferably, the water tank is made of high-transparency glass or acrylic material to ensure excellent optical performance and stability during the shooting process. The water tank is used to accommodate the experimental target and simulate the basic environment of the underwater scene, and its size is set according to the experimental requirements. Usually, the length direction of the water tank is greater than 7m, and the width and depth directions are greater than the size of the experimental target, so as to flexibly adjust the position of the camera during the experiment and simulate different scene depths and shooting angles.

[0008] Preferably, the light source device includes adjustable light sources with various brightnesses and spectra, which can simulate the lighting conditions in different underwater environments. The intensity and spectral range of the light source can be accurately adjusted according to the requirements. For example, white light for simulating shallow water areas and blue and green lights for deep water environments are used. The light source is equipped with diffuser sheets and filters, which can flexibly adjust the incident angle of light (such as 0°, 45°, 90°) and the uniformity of illumination, so as to truly reproduce the scattering and refraction effects of underwater light.

[0009] Preferably, the camera and its fixed bracket adopt a high-resolution camera with a resolution of not less than 4K to ensure that the captured images have high clarity and detail retention ability. The camera is installed on an adjustable bracket, and the bracket design supports flexible adjustment of various shooting angles (such as horizontal, top view, and elevation) and various shooting distances (such as 0.5m to 7m). The bracket structure is convenient for movement and stable locking, ensuring that the camera remains stable during the shooting process, so as to obtain highly consistent experimental data.

[0010] (2) Algorithm part

[0011] The algorithm designed by the present invention is used to control the experimental device to realize the automatic acquisition, processing, and storage of underwater images. The core functions of the algorithm include modules such as image parameter setting, image acquisition, image preprocessing, image cropping and pairing, image storage, and data annotation. Through a highly integrated modular design, the algorithm can capture high-quality clear images and degraded image pairs, ensuring the reproducibility and diversity of the dataset.

[0012] The described image parameter setting module is responsible for initializing the experimental parameters, including light conditions, camera position, and suspended particle concentration, etc. The light parameters support adjusting the brightness, spectral distribution (such as white light, blue light, green light), and incident angle (such as 0°, 45°, 90°) to simulate different underwater lighting environments. The camera parameters support adjusting the resolution (preferably not less than 4K), focal length, shooting distance (such as 0.5m to 7m), and angle (such as horizontal, top view, elevation). The suspended particle concentration is accurately set through a dosing device (such as 10g / m 3 —100g / m 3 ). All experimental parameters will be automatically recorded as metadata files for easy reproduction and analysis of subsequent experiments.

[0013] The described image acquisition module completes the acquisition of clear images and degraded images by controlling the experimental device. First, under the condition of no suspended particles added, clear images are acquired as reference images (Ground Truth). Subsequently, by adding suspended particles to the water tank and adjusting the light conditions, degraded images are acquired to simulate the real underwater degradation scenario. This module ensures that the shooting content of the clear images and the degraded images is consistent, and the files are automatically stored according to the number and label (such as image_01_clear.jpg and image_01_blurry.jpg).

[0014] The described image cropping and pairing module is used to unify the data format and generate paired samples of clear images and degraded images. First, the image cropping module crops the images according to the aspect ratio from 1:1 to 16:9, removes the background interference, and ensures that the target object is centered. Then, the clear images and the degraded images are aligned at the pixel level, and the redundant parts at the edges of the images are removed.

[0015] The described image saving module saves the processed images and their metadata in a standard format. The cropped images are paired by number and generate a unified naming (for example: image_01_clear.jpg and image_01_blurry.jpg), which is convenient for dataset management and subsequent applications. The clear images and the degraded images are classified and stored according to the experiment number and category, supporting common image formats (such as JPG, PNG). At the same time, the experimental parameters (such as light conditions, suspended particle concentration, camera distance, etc.) are saved as metadata files in JSON or CSV format and associated with the corresponding image data.

[0016] The present invention combines a physical device and an automated acquisition algorithm, and aims at the problems existing in the existing underwater image datasets, such as the reference image generation method relying on algorithms, being not real enough, and the degraded images not being close to the actual underwater environment, and proposes an underwater image dataset generation method based on real physical degradation. Compared with the prior art, the present invention has the following beneficial effects:

[0017] (1) Real physical degradation, accurate and reliable reference images: The present invention constructs a real physical degradation environment through device modules such as a transparent water tank, suspended particles, and light control. The degraded images are directly acquired and generated by physical means, avoiding problems such as color distortion, detail loss, or mapping distortion that may be caused by algorithm generation, thereby greatly improving the authenticity and reliability of the dataset, and providing high-quality and trustworthy basic data for subsequent underwater image enhancement and related research.

[0018] (2) Parameters are controllable and can be easily reproduced: By precisely setting and recording experimental parameters (including light conditions, suspended particle concentration, camera distance and angle, etc.), the present invention realizes the full-process controllability of the degradation environment. The experimental parameters are preferably stored in the form of metadata files, which is convenient for experimental reproduction and result comparison and analysis, providing standardized reference conditions for data generation and algorithm evaluation.

[0019] (3) High diversity, covering a wide range of degradation scenarios: By adjusting the suspended particle concentration, light intensity and spectral distribution, camera parameters, etc., the present invention can simulate a variety of underwater environments, including shallow water areas, deep sea areas, low-light environments, and different turbidity conditions. The generated dataset has high diversity, covering a variety of degradation types, significantly improving the applicability of the dataset.

[0020] (4) High efficiency and automation, improving data generation efficiency: The automatic acquisition algorithm of the present invention can control data acquisition, processing, pairing and saving throughout the process, greatly reducing manual intervention and significantly improving data generation efficiency. The cropping, pairing, annotation and expansion of data are all automatically completed by the algorithm, ensuring the generation of efficient and high-quality datasets.

[0021] (5) Strong scalability, adapting to various research needs: The present invention supports flexible experimental parameter settings and various data expansion methods, such as image enhancement (rotation, cropping, flipping, etc.) and format conversion (adapting to frameworks such as TensorFlow and PyTorch). In addition, this technology can be used to generate datasets suitable for various tasks, including underwater image enhancement, object detection, image segmentation, object tracking, etc., providing multi-purpose basic data for related research.

[0022] (6) Unified pairing structure, supporting supervised learning: The clear images and degraded images generated by the present invention are strictly paired and saved with a unified naming rule and structure, directly supporting the training and evaluation of supervised learning algorithms and facilitating the rapid use by algorithm developers. Brief Description of the Drawings

[0023] Figure 1 It is a schematic diagram of the overall structure of the image acquisition device of the present invention. In the figure: 1 - water tank, 2 - experimental target, 3 - vertical rod of the camera bracket, 4 - horizontal rod of the camera bracket, 5 - underwater camera, 6 - underwater fill light, 7 - power supply, 8 - computer display device, 9 - water body.

[0024] Figure 2 It is a schematic diagram of the flow of the method for generating the image dataset of the present invention. Detailed Embodiments

[0025] To make the technical means, creative features, achieved purposes and effects of the present invention easy to understand, the present invention will be further described below in conjunction with specific embodiments.

[0026] Example 1: Acquisition of Clear Images

[0027] The image acquisition device of the present invention is as Figure 1 shown, including: water tank 1, experimental target 2, vertical rod 3 of camera bracket and cross bar 4 of camera bracket, underwater camera 5, underwater supplementary light 6, power supply 7, computer display device 8, and water body 9.

[0028] Clean the water tank (1) to ensure that the experimental environment is pollution-free and has good transparency. Place the experimental target 2 in the center of the water tank 1, and fix the underwater camera 5 through the camera brackets 3 and 4. Adjust the camera brackets 3 and 4 to align the camera with the experimental target 2, and adjust the angle and position to ensure that the target object is centered and completely enters the shooting frame. Turn on the underwater supplementary light 6 to provide uniform lighting conditions, and at the same time, adjust the focus, exposure, and resolution of the camera in real time through the computer display device 8 until a clear and stable image is obtained. Turn on the camera 5 to acquire clear images of the experimental target, and the images are stored in the computer display device 8 in JPG format, and the files are named image_01_clear.jpg according to the number.

[0029] Example 2: Acquisition of Degraded Images

[0030] After acquiring clear images, degraded images with different degradation degrees are generated by adjusting the experimental environmental conditions. Specifically, it includes: adding suspended particles (such as sand grains, starch, or microplastic particles) into the water tank 1, stirring to make them evenly distributed, and controlling the concentration within the range of 10 g / m 3 to 100 g / m 3 (such as 20 g / m 3 , 60 g / m 3 , 100 g / m 3 ) to simulate water body environments with different turbidities; turn on the underwater supplementary light 6, and adjust the brightness, spectral distribution, and light incident angle (such as 0°, 45°, 90°) to simulate underwater lighting conditions at different depths and directions; adjust the distance (such as 0.5 m, 3 m, 7 m) and shooting angle (such as horizontal, top view, bottom view) of the camera 5 to acquire image data from multiple perspectives and distances. Under each set of environmental conditions, degraded images are acquired through the camera and stored in the computer display device 8 according to the number, for example, named image_01_blurry.jpg.

[0031] Example 3: Image Processing and Saving

[0032] After the acquisition is completed, first, the image cropping module crops the image with an aspect ratio of 1:1 to 16:9 (such as 4:3), removes the background interference at the edge of the water tank or non-target area, ensures that the experimental target is centered in the image, and at the same time removes the redundant parts at the edge of the image to improve the data quality. Then, the SIFT feature point matching alignment algorithm is used to accurately align the pixels of the clear image and the degraded image, ensuring that the two match in content and correcting the slight offsets caused by differences in shooting angles or environments. Finally, the clear image and the corresponding degraded image are paired according to a unified numbering rule, for example, named image_01_clear.jpg and image_01_blurry.jpg, and saved to the experimental data folder for subsequent data processing and analysis.

[0033] Example 4: Image Processing and Saving

[0034] The processed clear images and degraded images are stored in JPG format and organized according to numbering and folder classification. The folder names include the experimental date and parameter tags for efficient data management and quick retrieval. At the same time, the relevant parameters of each experiment (such as light conditions, turbidity, camera distance and angle, etc.) are recorded in JSON or CSV file format and attached to the corresponding image folder and saved together with the image data to ensure the integrity of the experimental information and provide reliable support for subsequent data analysis and experiment replication.

[0035] The above shows and describes the basic principles, main features and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited by the above embodiments. What is described in the above embodiments and the specification only illustrates the principles of the present invention. Without departing from the spirit and scope of the present invention, the present invention will have various changes and improvements, and these changes and improvements all fall within the scope of the present invention claimed. The scope of protection claimed by the present invention is defined by the appended claims and their equivalents.

Claims

1. A method for generating underwater image dataset based on real physical degradation, characterized in that: The following steps are involved: (1) Clear image acquisition: Place the experimental target in a transparent water tank and set the suspended particle concentration to ≤5 g / m 3 , Under the shooting conditions where the illumination uniformity is ≥90%, a clear image can be obtained by direct shooting; (2) Degraded image acquisition: By adjusting the experimental environment conditions, including suspended particle concentration, light intensity and spectral distribution, camera position and angle, various underwater degradation scenes are simulated and degraded images corresponding to clear images are acquired; (3) Image processing: The clear image and the degraded image are cropped with the experimental target as the center, with an aspect ratio of 1:1 to 16:9, and background interference is removed. The SIFT feature point matching algorithm is used for alignment to ensure pixel-level alignment between the two images, and redundant edge parts are removed. (4) Image pairing and storage: pair the clear image with the corresponding degraded image according to a unified numbering rule and store them in JPG or PNG format; (5) Parameter annotation and recording: Record the parameter information of each experiment, including light conditions, turbidity, camera distance and angle, and attach it to the image data in JSON or CSV file format.

2. The method according to claim 1, characterized in that: The clear image is obtained by the suspended particle concentration ≤ 5g / m 3 , and the illumination uniformity is ≥ 90%, and is used as the reference image for the degraded image.

3. The method according to claim 1, characterized in that: The degraded image is collected by the following steps: (1) Adjust the suspended particle concentration in the water tank to 10g / m 3 Up to 100g / m 3 , simulating underwater environments with different turbidity, suspended particles including but not limited to kaolin, milk, milk powder, and flour suspension; (2) Adjust the brightness and spectral range of light, including white light, blue light, and green light, and adjust the incident angle of light (0°, 45°, 90°); (3) Adjust the camera distance (0.5-7m) and shooting angle (horizontal, top view, top view) to obtain multi-view and multi-distance images.

4. The method according to claim 1, characterized in that: The image processing comprises: (1) Crop the image to remove the edge of the water tank and the background area to ensure that the experimental target is in the center of the image; (2) Using the SIFT feature point matching algorithm, the clear image and the degraded image are aligned and the redundant parts of the image edges are removed.

5. The method according to claim 1, characterized in that The image pairing and storage adopt a unified naming rule, the clear image is named image_number_clear.jpg, the degraded image is named image_number_blurry.jpg, and they are saved in folders according to the experimental date and parameters.

6. The method according to claim 1, characterized in that The parameter annotations include suspended particle concentration, light conditions, camera distance and angle, and are stored in JSON or CSV files and saved together with the image folder.

7. An apparatus for generating underwater image dataset based on real physical degradation, comprising: A transparent water tank (1) is used to contain a water body (9) and an experimental target (2), and can adjust the concentration of suspended particles in the water body to simulate an underwater environment with different turbidity; an underwater fill light (6) is used to provide adjustable light to simulate lighting conditions with different intensities and spectral distributions; an underwater camera (5) is installed on an adjustable bracket consisting of a camera bracket vertical rod (3) and a horizontal rod (4), and the bracket supports flexible adjustment of the camera distance and angle; a power supply (7) provides stable power for each component of the device; and a computer display device (8) is used to display and store the captured image data in real time.

8. The device according to claim 7, characterized in that The water tank is made of glass or acrylic material with a light transmittance of ≥95%, is greater than 7m in length, and is greater than the size of the experimental target (2) in width and depth, so as to facilitate adjustment of the camera position to simulate different scene depths.

9. The device according to claim 7, characterized in that The fill light can adjust the brightness and spectrum range of the light, and is equipped with a diffuser and a filter to adjust the uniformity and incident angle of the light.

10. The device according to claim 7, characterized in that The resolution of the camera is not less than 4K, the bracket supports a distance adjustment range of 0.5 meters to 7 meters, and can achieve horizontal shooting, top-down shooting, and upward shooting.

Images