Research On Underwater Image Enhancement And Object Detection Methods Based On Deep Learning

In recent years,with the continuous development of ocean exploration technology,the research of underwater images has gradually attracted the attention of many researchers.Underwater images contain abundant information,including underwater geographical environment,underwater biological distribution,and underwater mineral resources information.Underwater images have become an important source of information for researchers to study the underwater environment.To solve the problems of color deviation,detail blur,and texture distortion in original underwater images,this paper proposes a two-stage model SCDTNet that combines convolutional layers and Transformer blocks to improve the quality of underwater images.The proposed SCDTNet model considers the impact of shallow and deep features on image visual effects while taking into account the respective advantages of convolutional layers and Transformer blocks in processing different features.Corresponding module structures are designed to handle different features.For the shallow features of the image,convolutional blocks based on dense connections are used for processing to improve the details of the image and correct color deviation.Aiming at the deep features of images,a U-shaped structure processing based on channel processing Transformer blocks is designed to enhance the semantic information and deep texture of images.Experiments on open datasets have proven that the SCDTNet model has good underwater image enhancement effects.To solve the problems of low detection accuracy,missed detection,and error detection in current underwater target detection tasks,this paper proposes an improved model of Deformable-DETR-DA,which is used for underwater target detection tasks,based on the DETR-like detection models.The model considers the problem of insufficient feature extraction capabilities in the original Deformable-DETR model and designs a module that uses channel attention guidance to replace it.It also proposes and embeds an additional feature processing module to improve detection performance.Experiments on open datasets have demonstrated that the Deformable-DETR-DA model has good underwater object detection performance.To further improve the performance of the proposed underwater object detection model,this paper uses the proposed underwater image enhancement model SCDTNet to perform image enhancement preprocessing on the dataset,which improves the detection performance of the proposed Deformable-DETR-DA model.Experiments have proven the effectiveness of underwater image enhancement as a preprocessing for underwater object detection.