Developing machine learning methods for automatic recognition of fishing vessel behaviour in the Scomber japonicus fisheries

Introduction With a higher degree of automation, fishing vessels have gradually begun adopting monitoring method that combines human and electronic observers. However, the objective data systems (EMS) has not yet been fully applied in various boat scenarios such as ship behavior recognition. Methods In order to make full use EMS improve accuracy behaviors recognition vessels, present study proposes applying popular deep learning technologies convolutional neural network, long short-term memory, attention mechanism Chub mackerel (Scomber japonicus) vessel The operation process was divided into nine kinds behaviors, “pulling nets”, “putting “fish pick”, “reprint”, etc. According characteristics their work, four networks with different layers were designed pre-experiment. And feasibility each network observed. pre-experiment is optimized from perspective set network. From standpoint set, size significantly reduced, original are preserved much possible. combinations pooling, memory(LSTM) attention(including CBAM SE) added effects on training time effect compared. Results experimental results reveal methods outstanding performance vessels. LSTM SE module combination produced most apparent optimization model can achieve an F1 score 97.12% test surpassing classic ResNet, VGGNet, AlexNet. Discussion This research great significance management intelligent fishery promote development for ships.