Research On Underwater Image Enhancement And Underwater Biology Object Detection

With the increasing demand for underwater seafood,the research on underwater biology object detection technology can help robots replace human to complete the seafood fishing task.Therefore,the problem of high risk of artificial fishing operation can be improved,and it is beneficial to the development and utilization of marine resources.Underwater images are important carriers and forms of underwater information,and play a vital role in the exploration,development and utilization of marine resources.In recent years,deep learning technology has been widely used in underwater image enhancement and underwater biology object detection due to its powerful feature learning ability,and has achieved good results.However,these methods still have some shortcomings:underwater images are difficult to obtain in deep-sea environments,and it is also difficult for underwater images to have corresponding label values;the complex underwater environment and the absorption and scattering of light lead to low contrast and blurred details in underwater images,which seriously affect the perception and processing of underwater information;underwater biology objects are usually small and clustered,so details of underwater images cannot be fully extracted;in practical application,the underwater object detection model has large size and slow detection speed,which cannot meet the requirements of underwater real-time detection.In view of the above problems,the main research contents of this paper are as follows:1.This paper introduces the research status of underwater image enhancement and underwater biology object detection at home and abroad,which provides theoretical basis for subsequent experiments.The structure and principle of convolutional neural network,the basic theory of image enhancement and the method and principle of object detection are described in detail.The evaluation indexes of underwater image enhancement and underwater biology object detection are briefly introduced.It will pave the way for further research on underwater image enhancement and underwater biology object detection.2.Based on the problems of poor image quality such as low contrast and blurred details of underwater images,and insufficient local and detail enhancement of underwater images,this paper proposes a deep supervised residual dense network(Deep supervised residual dense network,DS＿RD＿Net),which is used to better learn the mapping relationship between clear in-air images and synthetic underwater degraded images.DS＿RD＿Net first uses residual dense blocks to extract features to enhance feature utilization;then,it adds residual path blocks between the encoder and decoder to reduce the semantic differences between the low-level features and high-level features;finally,it employs a deep supervision mechanism to guide network training to improve gradient propagation.Experiments results(PSNR was 36.2,SSIM was 96.5%,and UCIQE was0.53)demonstrated that the proposed method can fully retain the local details of the image while performing color restoration and defogging compared with other image enhancement methods,achieving good qualitative and quantitative effects.3.Due to the difficulty of underwater small object detection and the lightweight of underwater detection model,a lightweight underwater object detection method based on dense feature fusion(Bottleneck dense attention feature fusion yolov4-tiny model,BDAYOLOv4-tiny)is proposed in this paper,which can achieve the balance between accuracy and real-time of underwater object detection.In order to obtain more detailed image information,dense strategy was used to fully extract image features at different levels and scales.In order to improve the detection ability of small targets,three scales are detected simultaneously.Then the attentional feature fusion module(Attentional feature fusion,AFF)is added to better integrate semantic and scale-inconsistent features and improve the accuracy of multi-scale object recognition.Finally,a separable convolution module is introduced to replace traditional convolution for feature extraction,so as to reduce the number of parameters and improve the real-time detection performance.Experimental results show that the proposed method achieves good results on underwater object detection image datasets and video datasets,and achieves a good trade-off between underwater object detection speed and accuracy.