Coordinate-Aware Mask R-CNN with Group Normalization: A underwater marine animal instance segmentation framework

• DOI: 10.1016/j.neucom.2024.127488
• Publication Date: 2024-03-06T03:33:25Z
• Authors: Dewei Yi; Hasan Bayarov Ahm...; Shouyong Jiang; Yiren Li; Sean Joseph Flinn; Paul G. Fernandes
• Keywords: Normalization

Unsustainable fishing, driven by bycatch and discards, harms marine ecosystems. Addressing this, we propose a Coordinate-Aware Mask R-CNN (CAM-RCNN) method to enhance fish detection in commercial trawls. Leveraging CoordConv Group Normalization, our approach improves generalization stability. To tackle class imbalance, compound Dice cross-entropy loss is employed, image data are enhanced through multi-scale retinex color restoration. Evaluating on two fishing datasets, CAM-RCNN excels accuracy generalization, achieving the best Average Precision (AP) for instance mask BBOX prediction both source (39.7%, 40.2%) target domains (24.4%, 24.2%). This promotes sustainable selectively capturing desired fish, reducing harm non-target species.