Research On Algorithms Of Underwater Fish Detection And Species Recognition Based On Deep Learning

Marine fish is an important marine resource and an important source of protein intake for human beings.In the process of research and development of marine fish resources,it is necessary to identify the species of underwater fish and segment the image of fish.The accuracy of the algorithm is low,the generalization is poor,the segmentation effect is poor,and the segmentation position is inaccurate.With the rapid development of machine learning,the use of image enhancement and deep learning algorithms provides an effective way to solve the above problems.In this paper,the deep learning method is used to study the underwater fish recognition and image segmentation algorithm.The research results are as follows:（1）Aiming at the poor quality of the collected underwater fish images due to the harsh underwater environment and many interference factors,this paper proposes a fast image enhancement algorithm based on the improved ACE（Automatic Color Enhancement）algorithm.On the basis of the ACE algorithm,the computational complexity is reduced by a step-by-step quartet decomposition method,a smaller filter window is used to collect detailed information,and the gradually enhanced images are synthesized to obtain an image with the same size as the original.The simulation experiment results based on the underwater fish data set show that,compared with the ACE algorithm and other image enhancement algorithms,this method can effectively improve the quality of underwater images,and at the same time reduce the computational complexity of the image enhancement algorithm.Fish image segmentation and classification recognition have laid a good foundation.（2）In order to extract the feature information of deep underwater fish and make the segmentation position more accurate,this paper proposes the ARD（ACE-R-MCN-Deep）-PSPNet network model.This model is an underwater fish image segmentation algorithm based on the PSPNet（Pyramid Scene Parsing Ne twork）model.It uses the Res Net101 network as a feature extraction network,i ntroduces a depthwise separable convolution in the spatial pyramid pooling mo dule,and adds the R-MCN（Resnet Multi Convolutional Network）module.Impr ove the loss function to improve the loss function to make the model perform better.The experimental results show that the improved model has a significa nt improvement in all indicators.（3）There are interferences such as light,water patterns and complex backgrounds in the underwater environment,resulting in low recognition accuracy.In response to this problem,this paper proposes a fish recognition algorithm based on the PSA-Dense Net model,which is based on the Dense Net121 network model.Next,pre-training is performed by transfer learning,and the 3×3 convolution kernel in the tight block is replaced by the PSA（Pyramid Split Attention）module,and the loss function is improved to make the model converge faster.Through experiments on six underwater fish images,the experimental results show that the recognition accuracy of the improved model is significantly higher than other networks.（4）Under the Pycharm development environment,the underwater fish image enhancement,species recognition and segmentation software are designed through Py Qt5.The experimental results show that the software can realize the functions of image enhancement,image segmentation and species identification of underwater fish.