UAV-vision based Oilfield Anomaly Detection via Optimized LBAM-YOLO

Anomaly detection remains a great challenge for Oilfields. Oil leaks and intruding of unknown vehicles often propose severe safety issue for oil-field. At present, the using of Unmanned Aerial Vehicles (UAVs) is crucial in promptly identifying and addressing these anomalies, ensuring the safe operation of oilfields. However, detecting ground-level oilfield anomalies from a high-altitude is challenging due to the complexity and diversity of oil leakage features. Therefore, this study proposes an improved attention mechanism based target detection model termed LBAM-YOLO. The enhanced algorithmic model mainly incorporates the CBAM (Convolutional Block Attention Module) attention mechanism augmented with label information. Ablation experiments were conducted to evaluate the performance of our proposal. It demonstrates that compared to prevalent models, our method achieves more satisfactory performance in terms of recall and F1 score while maintaining high precision. Our model also exhibits more comprehensive ability to detect oilfield anomalies.