Abstract: The fish lateral line perception principle provides a new idea for the submarine underwater target sensing technology. Because the flow field model is difficult to build accurately, the model-based analytical method is difficult to precisely detect the target. The underwater translational target direction recognition method is proposed based on lateral line perception principle and deep learning. By constructing a pressure field distribution model around the dipole source, the lateral line perception principle is analyzed, and the relationship between pressure changes and the dipole source size, motion parameters, and position is theoretically analyzed. The results show that the pressure in the flow field is related to the position of the dipole source, and the features of pressure change at the dipole source vibration frequency are obvious, which can be used for training and identification. The time-frequency analysis method is used to process the pressure sensor signals and extract the time-frequency distribution features. Studies show that pressure changes in different translational directions have different time-frequency distribution features. It is proposed to use the convolutional neural networks to train the pressure sensor signals to identify the direction of the underwater translation target. The cross-shaped sensor array and the test platform are used to carry out test. Results show that the accuracy of the comprehensive recognition is above 80%. Without the need to accurately establish a flow field model, underwater translation targets can be accurately identified through deep learning. A new technical approach is provided for submarine underwater target perception.

Key words: bionic lateral line, deep learning, underwater translation direction recognition