Year: 2025
Type: conference-paper
Abstract: Sonar image targets clustering is an important research topic that requires further exploration. In this paper, to address the issue of poor clustering performance when applying clustering methods designed for optical images to sonar images, due to the significant differences between sonar and optical image targets as well as the uneven distribution of sonar image targets classes, an effective clustering method for sonar image targets is proposed based on the advanced Contrastive Clustering (CC) method. Firstly, Focal loss is introduced into the loss function of the instance-level contrastive head to mitigate the issue of uneven distribution of classes instances. Subsequently, the data augmentation method is redesigned to accommodate the characteristics of sonar image targets. The experimental results demonstrate significant improvements across all three metrics: Normalized Mutual Information (NMI), Accuracy (ACC), and Adjusted Rand Index (ARI), with maximum relative gains of 16% in NMI, 4% in ACC, and 16% in ARI.
Authors: Chaolong Liu, Juan Cheng, Shuai Shao
Language: English
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: gold
OpenAlex ID: W4411738868