Automated Fish Species Identification using Computer Vision

Indian National Centre for Ocean Information Services (INCOIS) provides Marine Fishery Advisory Services to the fishermen in India. Fisherman in India catches thousands of fishes everyday. Nevertheless, obtaining species-level catch reporting can be difficult for several reasons, including manual efforts that can lead to low or inaccurate reporting. This research suggests an automated fish identification and classification system that uses deep learning algorithms, notably YOLOv5 and ResNet50, to handle this problem. The dataset used in the research was manually created with images were taken from google. Dataset collected from Kaggle was also merged with the created dataset to increase the performance of the system. The system takes in input images of caught fish and accurately detects and classifies the fish species. The proposed system was tested on a dataset of several different species found in Indian subcontinent and the accuracy of 97 and 94 percent was achieved with YOLOv5 and ResNet50 respectively. This shows that YOLOv5 outperformed ResNet50. The results indicate that the proposed system can significantly reduce the time and effort required for species-level fish detection and classification, thus improving the accuracy of fishery advisory services.