Volatile organic compounds (VOCs) gases are common materials in chemical production, which contains various chemicals and has long-term effects on human health. Therefore, there is a need for safe and convenient detection of VOCs gases. Based on the characteristic of the infrared absorption peaks of VOCs gases, which are mainly distributed within a certain narrow wavelength, VOCs gases can be visualized through an infrared camera. However, infrared images have the characteristics of low resolution and low signal-to-noise ratio, and imaging enhancement will further extend noise in infrared images. To capture the difference between VOCs gases and disturbance, this paper proposes SYv5Net, which combines excellent representations of YOLOv5 and Swin Transformer in local features and global features respectively. In addition, SIoU is introduced in SYv5Net, which considers the location between groundtruth boxes and prediction boxes, and improves convergence speed during model training. Finally, based on publicly GasVid and the self-collected VOCs gas leakage infrared dataset, we integrate two datasets to create the final benchmark. Experimental results confirm that SYv5Net can effectively improve accuracy of VOCs gases detection in infrared videos and contribute to practical application.