Accurate Underwater ATR in Forward-Looking Sonar Imagery Using Deep Convolutional Neural Networks

Leilei Jin,Hong Liang,Changsheng Yang

doi:10.1109/access.2019.2939005

Abstract

Underwater automatic target recognition (ATR) is a challenging task for marine robots due to the complex environment. The existing recognition methods basically use hand-crafted features and classifiers to recognize targets, which are difficult to achieve ideal recognition accuracy. In this paper, we proposed a novel method to realize accurate multiclass underwater ATR by using forward-looking sonar—Echoscope and deep convolutional neural networks (DCNNs). A complete recognition process from data preprocessing to network training and image recognition was realized. Firstly, we established a real, measured Echoscope sonar image dataset. Inspired by the human visual attention mechanism, the suspected target region was extracted via the graph-based manifold ranking method in image preprocessing. Secondly, an end-to-end DCNNs model, named EchoNet, was designed for Echoscope sonar image feature extraction and recognition. Finally, a network training method based on transfer learning was developed to solve the problem of insufficient training data, and mini-batch gradient descent was used for network optimization. Experimental results demonstrated that our method can implement efficiently, and the recognition accuracy on a nine-class underwater ATR task reached 97.3%, outperforming traditional feature-based methods. The proposed method is expected to be a potential novel technology for the intelligent perception of autonomous underwater vehicles.

Highlights

Accurate target recognition is a crucial basis for underwater exploration and ocean development
ECHOSCOPE SONAR IMAGE DATASET We evaluate the effectiveness of our underwater target recognition method on a real, measured sonar image dataset
For our designed deep convolutional neural networks (DCNNs), one can see that NL = 2 gives the best result, which suggests that completely fine-tuning the target network (NL = 0) may not get the best performance

Summary

Introduction

Accurate target recognition is a crucial basis for underwater exploration and ocean development. Due to the changeable environment and limitations on the sensing of the marine, accurate multiclass underwater automatic target recognition (ATR) has not been commendably solved. We choose the type of imaging sonar according to the practical application. Sidescan sonar is a system used for generating an image of the sea bottom area, which is not suitable for the identification of floating objects, nor for the real-time recognition tasks. The main difficulty with SAS relates to micro navigation and platform trajectory estimation, the high requirement on the platform movement limits its application scenario. Realtime imaging sonar Echoscope is one of the most important innovations in underwater observation in recent years. The invention of a more delicate and efficient imaging system facilitates the generation of high-resolution images and the design of appropriate technologies to automatically

Results

Discussion

Conclusion