Fine-grained Classification Of Fish Based On Deep Learning

Fishery oversight has long been a key issue both domestically and internationally.Traditional methods of monitoring fishing operations often rely on national laws and the establishment of regional and global fisheries conservation organizations.These methods have several drawbacks,including poor monitoring,high labor cost,and low productivity.In order to realize the supervision of fishery activities by identifying and classifying the fish caught by fishing vessels,a fine-grained classification method based on deep learning was proposed.The sustainable development of Marine resources and the fishing industry can greatly benefit from this research in both economic and scientific terms.The following are the main findings of this study:(1)We divide the classic fine-grained fish classification problem into two tasks:image enhancement and image classification,in accordance with the features of fish data.Studies reveal that the performance of the network is significantly enhanced by combining our suggested image enhancement network and image classification network.(2)We use a super resolution adversarial generation network for the image improvement network.Through testing,it was discovered that the super resolution adversarial generation network that we trained can enhance the classification performance of deep learning networks by significantly improving the sense of reality and clarity of images by reducing image noise and enhancing detail information.At the same time,it can also produce positive effects when utilized for data augmentation.(3)We select Sawin-Transformer and Efficient Net,two popular approaches,as our basis network for the image classification network since they excel in tackling fine-grained classification problems.The loss function,attention mechanism,and image enhancement network were used to improve the fish fine-grained classification network.Finally,we suggest three networks for fine-grained fish categorization based on deep learning.(4)We install the fine-grained fish classification network that has been trained on the mobile terminal and test it in the real-world environment.The final test result complies with the desired outcome.