Research On Underwater Biometric Identification Method

Underwater aquaculture,ocean mapping,and subsea engineering applications are difficult to carry out and need significant financial resources due to the complexity of the marine environment.Artificial intelligence development offers a fresh approach to resolving the issues that science and technology are currently facing.In addition to ensuring the safety of people performing these risky operations,using underwater robots for visual assistance lowers costs.Nowadays,acoustic and optical signals are the primary methods used by underwater vehicles to acquire targets.The optical signal,which is utilized frequently in the field of ocean detection,has the best ability of these to express the target and collect more accurate information.Despite the visual benefits of images collection,the underwater environment greatly compromises vision.The image data that the sensor receives is hazy as a result of the light’s attenuation and dispersion in the water.Thus,underwater biometrics are challenging to implement.The underwater ecology is home to a variety of organism types,which causes issues with classification and recognition,including low accuracy and poor generalization.This results in issues like limited generality and accuracy.The following research accomplishments can be attributed to the use of pertinent methods and models to augment and restore biometric features and recognition tasks of underwater images as a result of the rapid development of non-deep learning techniques and deep learning models:(1)The underwater image is improved and restored using the fusion approach of physical model and Laplacian image in order to address the issue of underwater image color distortion caused by stray particles,plankton,and other causes.Underwater imaging model restoration is the foundation for physical model restoration,which can more correctly recreate the underwater scene.Red channel information shortage is resolved using Laplacian image fusion.UCIQE and PSNR are improved by the technique by 0.0262 and 2.0444 respectively.The total average decrease of NIQE is 0.5737,showing that this approach may improve the color saturation of underwater images with severe red channel absence and increase the visibility of the biological contour features.(2)An underwater image enhancement system with multi-scale semantic features is suggested to address the issues of poor contrast and poor color saturation of underwater degraded images.The technique uses the encoding and decoding structure as primary framework,using the multi-scale semantic feature module to integrate data from various scales and enhance the network’s capacity for semantic information perception.The generation module and the multi-scale fusion module both incorporate the attention mechanism at the same time to improve the network’s capacity to extract various pixel attributes.Also,a Gaussian pyramid function is created by fusing the global similarity,WGAN,and VGG19 color perception loss functions in order to improve the image.With this approach,the UCIQE and PSNR are elevated by 0.0518 and 2.5358,respectively.The overall average decrease of NIQE is 0.3092,showing that this method can successfully differentiate the bright and dark portions of images and restore the real color of complicated underwater images and the biological details are more obvious.(3)The YOLOv5 s enhanced model was employed for underwater biological recognition in order to address the issues of low accuracy and sparse data set labeling in the classification and recognition of underwater biological species.The Vi T-Transformer model and CSPCBAM model are introduced to this approach to improve the feature extraction capability,resulting in a recognition accuracy of 0.9740,and this algorithm is also confirmed that the improved image technology helps to increase the reliability of underwater biometrics.