Optimization of deep convolutional neural network using Bayesian method to increase the detection rate of diving diver with open circuit system

The detection of the diver signature that diving with an open circuit system is important for several reasons, including identifying hostile divers in case of intrusion into underwater infrastructure, warning of divers entering into the aquaculture areas, preventing their collision with vessels, and instantaneous monitoring of their operations. In this paper, a deep convolutional neural network optimized by Bayesian method is used to detect divers. In order to adapt the problem to the intended deep network, after collecting the required data, bayesian optimizer search method is used to fine-tune important hyperparameters. After achieving the optimal values of important hyperparameters, three models of Bayesian, Alexnet and Darknet19 are used to classify signals in the form of two classes of Diver and No-Diver; as the results show, Bayesian search method, in addition to increasing the accuracy of training model, significantly saves calculation time in accordance to the benchmark models. So that in a total time of 1772 seconds the average square error of 0.00091977 is calculated.