Side-scan sonar image feature extraction method based on full convolutional neural network

Abstract

The invention provides a side-scan sonar image feature extraction method based on a full convolutional neural network. The method comprises the following steps: carrying out the data augmentation through an original sonar image, and set needed by model training and testing; manually labeling the edge area of the submarine topography of each image in the sample set, distinguishing a target and a background, and obtaining a model training and testing label graph; an FCNs model is constructed; inputting the submarine topographic map and the corresponding label map into a network, training the network by adopting a small-batch gradient descent method of a driving quantity item, and storing an optimal network model; comparing convergence and stability of the network under a random gradient descent method and a small-batch gradient descent method; and extracting terrain edge contour features, outputting a feature extraction result, and carrying out qualitative evaluation on the result. According to the method, complex preprocessing is not needed, and the sonar feature extraction method is high in speed, high in efficiency and high in speckle noise resistance; the performance of the network is improved, and the convergence and stability of each network model of the FCNs are ensured.

Description

technical field [0001] The invention belongs to the field of feature extraction of sonar images, and in particular relates to a method for extracting features of side-scan sonar images based on a fully convolutional neural network. Background technique [0002] Due to the particularity of the marine environment, sonar is the most effective sensor for underwater detection compared to optical photography in the detection of seabed targets, the exploration of marine mineral resources, and ocean surveys. The side-scan sonar system was born in the late 1950s, and has been widely used in the tracking and identification of underwater military targets, the exploration and development of seabed mineral resources, the automatic drawing of seabed topographic maps, fish detection and seabed natural Environmental investigations and many other fields. In order to complete specific tasks accurately and efficiently, it is often necessary to perform feature extraction on targets in sonar im...

AI Technical Summary

Problems solved by technology

[0006] The object of the present invention is to provide a kind of side-scan sonar image feature extraction method based on full convolutional neural network, overcome the slow speed of traditional manual featur

Images