Development of a Detection System for Endangered Mammals in Negros Island, Philippines Using YOLOv5n
Angelo De Castro, Michael Aaron Sy, John Alfred J. Castañeda, Myles Joshua Toledo Tan, Nouar AlDahoul, Hezerul Bin Abdul Karim · 2022
Many species have gone extinct as a result of human neglect and various environmental influences. Monitoring these species has proven to be a challenge due to their small population, remote habitats, and evasiveness, among other reasons. Nonetheless, they can be routinely tracked by using CCTV cameras. This project made use of a transfer learning approach to detect the presence of Malayan civets (Viverra tangalunga), Visayan leopard cats (Prionailurus javanensis sumatranus), Visayan spotted deer (Rusa alfredi) and Visayan warty pigs (Sus cebifrons) in the forests of Negros Island. We also developed a web application to record camera data and assign timestamps. The setup consisted of a Raspberry Pi 3B+ and a Raspberry Pi camera module powered by a solar power bank. The method used YOLOv5n, a lightweight object detection algorithm, to detect the four species. The trained model yielded 91% mAP, 64% [email protected]:0.95, and the following average precisions: 94% (Visayan warty pig), 91% (Malayan civet), 88% (Visayan leopard cat), and 91% (Visayan spotted deer) with 4.5 GFLOPs.