Zero-sample side-scan sonar image target classification method

Abstract

The invention discloses a zero-sample side-scan sonar image target classification method, and the method comprises the steps: obtaining conventional optical images of the same type according to the target type information of any to-be-recognized side-scan sonar image, carrying out type marking of the conventional optical images to acquire some side-scan sonar images which are easier to obtain anddo not contain to-be-recognized types of targets; using style migration method for the side-scan sonar image with a conventional optical image and a side-scan sonar image as input; generating a simulated side-scan sonar image of a certain specific category, training a deep neural network according to the generated simulated side-scan sonar image data set, and finally, using the deep neural networkobtained by training the simulated side-scan sonar data set to be applied to classification and identification of the side-scan sonar image of the certain specific category. The side-scan sonar imageclassification of a specific category of targets can still be accurately realized without available training samples, and the problem that a recognition network cannot be trained due to the fact thatthe samples cannot be obtained is solved.

Description

technical field [0001] The invention relates to a zero-sample side-scan sonar image target classification method, in particular to a zero-sample side-scan sonar image target classification method using pseudo-sample synthesis. Background technique [0002] Side-scan sonar is the most widely used sensor in underwater detection, target search and other fields. In recent years, with the development of ocean development activities, the application of sonar equipment is not limited to military applications, but also has a wide range of applications in commercial and civilian fields. , such as seabed resource detection, oil exploration, sea rescue, automatic drawing of seabed topography and landform maps, and fish detection, etc. [0003] However, most of the current underwater robots only have the function of acquiring side-scan sonar image data, and the interpretation and interpretation of side-scan sonar images are still mainly done manually. When the sea area to be scanned is...

AI Technical Summary

Problems solved by technology

[0006] In view of the above-mentioned prior art, the technical problem to be solved by the present invention is to provide a zero-sample side-scan sonar image target classification method using pseudo-sample synthesis, through conventional optical images of the category to be identified and some other types of side-scan sonar images, Synthesize the simulated side-scan sonar images of the class to be identified, thereby converting the zero-shot problem into a supervised learning problem, thereby overcoming the inability to apply supervised learning due to the lack of available training samples

Images