Underwater acoustic data set augmentation method and system based on conditional generative adversarial network

Abstract

The invention discloses an underwater acoustic dataset augmentation method and system based on a conditional generative adversarial network, and the method comprises the steps: 1, obtaining the slices of a target object in a sonar dataset and a remote sensing dataset, carrying out the contour enhancement and pixel filling of a target object image in the slices, retaining the class information of the target object in the qualified slices, and carrying out the recognition of the target object; classifying and storing as samples; step 2, through a conditional generative adversarial network, augmenting the samples which do not reach a preset number threshold in the step 1; and step 3, pasting the samples reaching a preset number in the step 1 and the samples augmented in the step 2 to the seabed reverberation background image in the sonar data set, performing optimization processing on the pasted image, simulating the image into a sonar image, and forming the sonar data set. According to the method, the underwater acoustic image data set is enriched, and the development and application of the deep learning method in underwater target detection and segmentation tasks are promoted on the basis, so that the accuracy of environmental perception target detection is improved.

Description

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine

Login to view more

AI Technical Summary

Benefits of technology

This patented technology helps gather more detailed sound waves from an ocean bottom by analyzing them for targets or objects that are hidden behind it. It can help identify these areas with high precision based on their depths within water.

Problems solved by technology

This patents describes various technical challenges related to improving the performance of undersea soundings detectors such as depth cameras or radars. Current techniques require manually annotated training datasets from scratches over water, making them expensive and limited in their applicability within complicated environmental settings where many different kinds of objects may exist together. Additionally, these conventional approaches often result in insufficiently diverse sample sizes leading to lower accuracy levels compared to more advanced models like convolutional neural networks trained through machine learning.

Images