Oilfield Staff Workwear Detection and Monitoring System

In this paper, deep learning algorithms are applied to oilfield work scenarios, and YOLOv5s is selected as the backbone network for staff workwear detection and analysis of existing problems: To improve the small-scale objective detection capability, the second layer of the backbone network is added for feature output and the SimAM attention mechanism module is introduced. To improve the obscuration objective detection capability, the G-Transformer encoder module is introduced, and the Weighted boxes fusion (WBF) candidate frame screening method is selected for post-processing. After that, the above-improved method is designed in general, and the G-YOLOv5s detection network is proposed. Experiments show that G-YOLOv5s improves detection accuracy by 5.31% compared to the initial network, while the speed is only reduced by 20 frames/second. Finally, the PyQt tool with “OpenCV library + frame loop” as the core control method for monitoring system development can realize real-time monitoring of workers’ dressing in oilfield scenes, which has high practical value.