Reliable unmanned aerial vehicle-based thermal infrared target detection method for monitoring Procapra przewalskii in Qinghai
Procapra przewalskii plays a vital role in maintaining ecological balance; however, it faces considerable threats due to habitat degradation and illegal poaching. Monitoring this species using unmanned aerial vehicles (UAVs) has proven to be an effective conservation strategy. A major challenge in U...
Saved in:
| Main Authors: | , , , , , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
Elsevier
2025-12-01
|
| Series: | Ecological Informatics |
| Subjects: | |
| Online Access: | http://www.sciencedirect.com/science/article/pii/S1574954125002183 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | Procapra przewalskii plays a vital role in maintaining ecological balance; however, it faces considerable threats due to habitat degradation and illegal poaching. Monitoring this species using unmanned aerial vehicles (UAVs) has proven to be an effective conservation strategy. A major challenge in UAV-based surveillance of Procapra przewalskii is conducting observations at night or under conditions of poor visible light. To address this issue, this paper presents a thermal infrared (TIR) target monitoring technique using UAVs. This technique employs YOLOv8s as the base model and proposes a multi-frame processing (MFP) method (YOLO-MFP). This method uses the current frame as the primary input and combines optical flow–processed images and background-suppressed images as auxiliary inputs. Background-suppressed images can effectively minimize most background pixels, while regions with high vector values in optical flow–processed images indicate object positions. The model extracts raw feature data, object details, and movement information from these inputs to improve detection performance. Additionally, a small target detection layer is added to reduce missed detections of smaller targets in TIR images while enhancing the overall detection accuracy. Furthermore, the VoVGSCSP module refines the model's neck architecture by effectively merging the feature maps across various stages, reducing computational demands without sacrificing detection precision. Finally, through numerous comparative experiments on our proposed TIR-Procapra przewalskii dataset, YOLO-MFP reaches a mean average precision (mAP@0.5) value of 96.4 %, precision value of 92.6 %, and recall of 97.0 %, making it superior to the current state-of-the-art models. The importance of this study lies in its enhanced monitoring capabilities for Procapra przewalskii, providing valuable insights for future UAV-based wildlife observation efforts. |
|---|---|
| ISSN: | 1574-9541 |