Marine target detection method and system based on self-supervised representation learning

Abstract

In the marine target detection method and system based on self-supervised representation learning, in the self-supervised ship feature learning stage, the momentum comparison method is used to train the feature extraction model on the unlabeled marine target data. The momentum comparison process maintains the dictionary as a sample queue. And update the key encoder by using the momentum update method; in the supervised ocean target detection stage, the Faster R-CNN model is used for ocean target detection, and the Faster R-CNN model includes a backbone network for feature extraction and a region of interest generated Region candidate network and RoI Head network; through the backbone network, the feature extraction network parameters obtained in the self-supervised ship feature learning stage are used to initialize, to provide prior knowledge of the marine environment and ships for the target detection model, and to train the backbone network while training the model Parameter fine-tuning. The invention reduces the impact of insufficient ocean data samples on the detection effect.

Description

technical field [0001] The invention belongs to the technical field of marine target detection, in particular to a marine target detection method and system based on self-supervised representation learning. Background technique [0002] As a major marine country with vast seas, marine strength is an important part of my country's comprehensive national strength. Improving the strength of marine science and technology is of great significance to building a strong marine country. Real-time monitoring of sea areas with the help of unmanned boats and other maritime unmanned equipment can effectively strengthen sea area control and maintain my country's marine security. Therefore, how to improve the intelligent perception capability of maritime unmanned equipment has become one of the key issues in the field of marine science and technology. Object detection has also become one of the hot research directions in the field of marine environment perception. [0003] As one of the mo...

AI Technical Summary

Problems solved by technology

[0006] To this end, the present invention provides a marine target detection method and system based on self-supervised characterization learning, which solves the problem of limited application of deep learning algorithms in the field of marine target detection caused by the lack of large-scale marine target data sets, and makes full use of label-free Using marine data as the starting point, self-supervised representation learning is introduced into marine target detection to reduce the impact of insufficient marine data samples on detection results

Images