Joint Multi-Scale Channel Attention and Multi-Perception Head for Underwater Object Detection

Today's article comes from the Journal of Big Data Mining and Analytics. The authors are Shi et al., from Southwest Jiaotong University, in China. In this paper, they're building an object detection pipeline that can operate even in murky, turbid water. Their system, JMM-Net, introduces two components: a Multi-Scale Channel Attention module that strengthens a CV model's sensitivity to subtle color and texture cues, and a Multi-Perception Head that improves how it recognizes small, faint, or overlapping objects.

DOI: 10.26599/BDMA.2025.9020030