Selection of Training Options for Deep Learning Neural Network Using Genetic Algorithm

Recently, a growing usage and consequently a developing level of autonomy of Autonomous Underwater Vehicles (AUVs) can be seen. These vehicles are power supplied and controlled from the sources located on their boards. One of the most often used sensors of the AUV is a video camera. This sensor in connection with the video images processing software can increase the level of autonomy of the AUV. One of the most popular applications using video camera is an image recognition, e.g. for the obstacle detection. One of the newest methods used for this application is the Deep Learning Neural Network (DLNN). The goal of the paper is to examine the genetic algorithm optimization method for the selection of training options for DLNN used for the underwater images recognition. In the research, the pretrained AlexNet DLNN and the Stochastic Gradient Descent with Momentum (SGDM) training method have been used. It is planned to implement examined DLNN on board of the Biomimetic Underwater Vehicles (BUV)