MoDA-TL - monitoring domestic animals using convolutional neural networks and transfer learning

In recent years, computer vision has made significant advances, expanding its knowledge and applications in various fields. An important example is the use of this technology to improve the recognition of different types of animals. This paper proposes an intelligent surveillance system that can individually identify each animal in a specific location and clearly indicate dangerous or unsuitable areas during monitoring, ensuring the safety of both people and the animals being monitored. In this context, deep learning algorithms, such as convolutional neural networks (CNN), are used to produce machine learning models capable of detecting and identifying objects in digital images. The study utilises the you only look once (YOLO) version 8 model and achieves 99.5% accuracy in animal recognition, demonstrating its effectiveness in monitoring. Additionally, a comparison between a model trained from random weight initialisation and another based on transfer learning reveals that the latter outperforms across various metrics, showing 99.5% accuracy, 99.3% recall, 99.5% mAP50, and 77.5% mAP50-95. These results highlight the advantage of transfer learning in optimising performance.