Detection of Fish Species Captured by Artisanal Canoes Using YOLOv11

In Senegal, fish are caught and classified manually, and declarations are often incomplete. Data collection at fishing wharves on the Senegalese coast is limited by the lack of suitable tools for identifying and effectively recording catches. Although studies have been carried out to improve management in the fisheries sector, there is a notable lack of local databases and automated systems for detecting and counting fish. This article proposes a detection and identification model, which should contribute to the creation of a suitable local database. The data used are images taken on target docks and the data are enriched with information from external sources. These data were used to produce a segmented, structured set of views. The database is used with YOLO v8 and YOLO v11 to improve the model. Image detection using bounding boxes is a key step in training the model. It is trained a priori by YOLO v8, which produced recall and confidence scores ranging from 0.01 to 0.75. These results are enhanced in this paper by the YOLO v11 model. Very efficient for automatic detection and counting, this model shows promising performance for simulations. Metrics such as mAP50 (with an IoU threshold of 0.5) and mAP50-95 (average over several thresholds) show a clear evolution, indicating a significant progress in the detection capacity of the model. These results highlight the accuracy and efficiency of YOLO v11 in identifying and positioning bounding boxes.