AUV Path Planning and Image Recognition Based on Convolutional Neural Network

摘要

Autonomous Underwater Vehicle (AUV) has been widely used in marine resources development and underwater engineering operations. Vision is one of the most important environmental testing technologies at present, and Convolutional Neural Networks are widely used in target testing. Based on LENET-5, a Convolutional Neural Network (CNN) for underwater training set was proposed, and image recognition was implemented through Raspberry PI. AUV controls the camera and underwater lighting through Raspberry Pi, and builds a model based on the TensorFlow framework. The tested dataset showed that the accuracy of the 7-layer Convolutional Neural Network is only 80%. So,in this paper, the Convolutional Neural Network was improved to finally design a 13-layer neural network, including 1 input layer, 4 convolutional layers, 4 pooling layers, 3 fully connected layers, and 1 output layer. After 100 iteration training experiments, the accuracy of underwater target recognition reached 99.18%, and the independent underwater robot underwater target recognition was achieved.