A Comparison of YOLO and Mask R-CNN for Segmenting Head and Tail of Fish
Eko Prasetyo, Nanik Suciati, Chastine Fatichah
Abstract
The visual appearance of the fish’s head and tail can be used to identify its freshness. A segmentation method that can well isolate those certain parts from a fish body is required for further analysis in a system for detecting fish freshness automatically. In this research, we investigated the performance of two CNN-based segmentation methods, namely YOLO and Mask R-CNN, for separating the head and tail of fish. We retrained the YOLO and Mask R-CNN pre-trained models on the Fish-gres dataset consisting of images with high variability in the background, illumination, and overlapping objects. The experiment on 200 images containing 724 heads and 585 tails annotated manually indicated that both models work optimally. YOLO’s performance was slightly better than Mask R-CNN, shown by 98.96% and 96.73% precision, and 80.93% and 75.43% recall, respectively. The experimental result also revealed that YOLO outperforms Mask R-CNN with mAP of 80.12% and 73.39%, respectively.