Detection of gas leaks in petrochemical plants often involves installation cameras to transmit images to a control room for monitoring. However, the potential for disasters arises if monitoring staff fail to observe gas leaks promptly due to factors such as fatigue. Furthermore, some chemical plants utilize optical flame cameras to detect for flames due to gas leak through pipelines. However, it would be too late once the flames have been detected. Therefore, a real-time recognition of smoke generation due to leakage with alarm system is necessary. In the present study, an algorithm with artificial intelligence image recognition techniques has been developed to instantly detect and identify smoke leakage for petrochemical plants. A flame detecting fixed-camera is included in the monitor system with the developed algorithm for real-time gas leak detection and identification. It should be noted that the lack of actual gas leak smoke images on-site poses challenges for training deep learning models. To address this issue, the present study proposes the synthesis of virtual smoke with factory background images to supplement the insufficient real leak images.