Underwater Image Enhancement Algorithm Research For Marine Biological Recognition

The recognition algorithm of the marine environment is often ineffective.The strong absorption and the scattering of suspended particles interfere with the light,causing the captured underwater images to have defects such as color shift,image blur,low contrast,and low visibility.The severely degraded underwater image will blur target details and lose effective information for target detection,thereby increasing the difficulty of underwater operations.Therefore,it is necessary to enhance the image.This paper proposes an algorithm for underwater image enhancement,which uses convolutional neural networks and introduces jump connections to realize the fine-grained reconstruction of underwater low-quality images into clear images,so as to improve the detection accuracy of marine biological recognition.The main contents of this paper are as follows:Aiming at the characteristics of underwater images such as color distortion,image blur,low contrast,etc.,based on the convolutional neural network,this paper proposes an algorithm to reconstruct the original underwater image into a clear image through the network.The algorithm constructs a structure that includes image pre-processing coding,feature map enhancement reconstruction,and image post-image recovery decoding.Among them,jump connections are introduced to accelerate network convergence and prevent gradient explosion,and feature fusion is performed through concat to improve network performance..The algorithm is optimized according to the resource constraints of the embedded platform,and the visual effect of the reconstructed enhanced image is significantly improved.In view of the current shortage of paired data sets specifically used for underwater image enhancement,this paper uses CycleGAN's style transfer underwater image synthesis method to degrade clear marine life pictures on the one hand,and uses non-deep learning methods on the other hand.The poor real underwater pictures are enhanced,and finally a pair of clear-degraded marine biological image enhancement data sets are obtained.Finally,this paper designs a comparative experiment to analyze the enhancement effect of the algorithm in this paper.Experiments show that the algorithm in this paper has significantly improved the subjective visual effects,and achieved better results in image quality evaluation indicators.At the same time,the detection rate of marine biometrics has also been greatly improved,and the computing speed on the embedded platform is also Has obvious advantages.