Detecção de mamíferos em ambientes de mata utilizando YOLO

Abstract

The Earth’s biodiversity is increasingly threatened by various environmental challenges. To monitor and quantify biodiversity, biologists often use camera traps that record wildlife in their natural habitats, capturing videos when movement is detected within the camera’s range. However, these recordings frequently result in false positives, as factors like wind can trigger motion detection. In recent years, the use of intelligent systems for object detection in images and videos has gained significant attention, with machine learning and deep learning techniques showing strong performance across multiple fields. This study leverages deep learning, specifically using versions 3, 5, and 9 of the YOLO convolutional neural network, to detect animals in forest environments. Among these, the YOLOv5 model, with a batch size of 32, achieved the best results, boasting an F1-Score of 96.48%. Additionally, an application was developed to make these models more accessible to biologists.