Gas relay is an important protection device for oil-immersed transformers, which triggers alarm signals and generates protective actions by collecting decomposing gas in oil, but it is difficult to dynamically grasp the gas accumulation capacity and requires manual participation in subsequent processing. In this paper, a dynamic identification and automatic detection method of gas volume based on image analysis is proposed, which automatically introduces gas into the gas sensor system for characteristic gas composition analysis, so as to monitor the operating status of transformers in real time, check potential safety hazards and improve reliability. Firstly, an endoscopic camera is installed in the gas relay observation window to obtain oil level images, and the oil level position is detected by image analysis algorithm. Secondly, the nonlinear mapping relationship between oil level and gas capacity is established through BP neural network, so as to calculate the gas capacity. The experimental results show that the detection error rate of this method is only 3.61%, and the prediction error of most test samples is within 2%, which has high accuracy and can reliably realize the automatic protection of oil-immersed transformers.