A novel approach combining YOLO and DeepSORT for detecting and counting live fish in natural environments through video

Abstract

Applying Artificial Intelligence (AI) to the monitoring of live fish in natural environments represents a promising approach to the sustainable management of aquatic resources. Detecting and counting fish in water through video analysis is crucial for fish population statistics. This study employs AI algorithms, specifically YOLOv10 (You Only Look Once version 10) for identifying the presence fish in video frames, combined with the DeepSORT (Deep Simple Online and Realtime Tracking) algorithm to count the number of fish individual moving across the frames. A total of 9,002 frames were extracted from 13 videos recorded in five different environments: areas with submerged tree roots, shallow marine regions, coral reefs, bleached coral reefs and seagrass meadows. To train the recognition model, the dataset was divided into training, validation and testing sets in 8:1:1 ratio. The results demonstrated that the model achieved an accuracy of 89.5%, with processing times of 6.2ms for preprocessing, 387.0ms for inference and 0.9ms for postprocessing per image. The combination of YOLO and DeepSORT enhances the accuracy of tracking objects in aquatic environments, showing great potential for the monitoring of fishery resources.