Fish Population Status Detection Based on Deep Learning System

The detection of fish behavior is of great significance in aquaculture, and it is an important basis for judging whether the fish are normal or not. The timely detection and treatment of abnormal conditions will help to reduce the cost of aquaculture. At present, the technology is mostly focused on the recognition and detection of individual targets. This study attempts to detect the different states of moving objects of the same group, using convolution neural network, to achieve the purpose of qualitative analysis of group behavior. In this study, zebrafish are used as experimental subjects to detect the population behavior of fish. Through the expansion of existing data sets and the establishment of our own data sets to train the network, the CNN network based on VGG-16 is used for experiments, and relatively considerable experimental results are obtained. In the same sample state detection, compared with the ordinary texture detection, the accuracy of the experiment is improved by 5% by using the deep learning model, which has certain significance for guiding aquaculture and qualitative analysis of fish behavior.