81 / 2025-04-17 00:11:57

UAV based nimal detection for smart livestock farming

Animal detection,UAV,remote sensing,Livestock farming,attention mechanism

3. 智慧牧场与动物福利 > （1）智慧养殖的利惠：提高动物福利与健康、环境、经济效益与劳动力

Automatic animal detection and remote monitoring is a promising solution for vast and isolated farmlands or cattle stations in the modern livestock industry. The advancement in sensor technology, deep learning and the rise of Unmanned Aerial Vehicles (UAVs) have paved the way for farm management. In this work, we propose a deep learning based animal detection approach using UAV images and an improved YOLOv5 based on attention mechanism (YOLOv5-SE), aiming to achieve the best trade-off between detection accuracy and speed. The proposed YOLOv5-SE consists of two components: YOLOv5 extracts multi-scale features from each aerial image, while the attention mechanism SE block highlights the animal biometric-related features to enhance the animal detection performance. A challenging animal dataset was acquired to test this, consisting of three different animal species (i.e. cattle, sheep and dog) with complex backgrounds (e.g. illumination, shadows, and vegetation covering). Experimental results show that the proposed YOLOv5-SE achieved an detection precision of 94.48\% and a mAP@0.5 of 86.79% with inference occurring at 90.1 frames per second. Especially, the proposed YOLOv5-SE achieved 98.74% and 99.07% accuracies for sheep and cattle detection, respectively. All these yielded values are superior to state-of-the-art methods, including Faster R-CNN, RetinaNet, YOLOv4 and YOLOv5. In addition, the influence of different spatial resolution on animal detection performance was investigated. Experimental results show that spatial resolution higher than 0.5 m/pixel and network input size of 416×416 is favorable for real-time animal detection and monitoring. Overall, the proposed YOLOv5-SE based approach could effectively capture key animal biometric features and enhancing the detection performance. Our research shows promising steps toward the incorporation of artificial intelligence with UAV data, for remote animal detection in smart livestock farming.