As a high?risk area, the fire safety has always attracted much attention. Although smoke and flame alarm have been widely used, there are still problems such as single ?point detection and easy environmental impact. In response to such problems, a multi?way flame smart video monitoring system based on the B/S architecture is designed and implemented, and it is presented in the form of a web system. In the system, an improved YOLOV5 flame detection algorithm is integrated. The Ghost convolution is used to replace in conventional convolution to achieve the lightweight of the network, and the improved attention mechanism modules and small target detection anchor frame is added to enhance small target detection ability. Finally, the flame movement information extracted from the Optical flow network and the original flame data is sent into the improved YOLOV5 flame detection algorithm to further improve the detection accuracy of the flame. A large number of on?site test proves that the system can identify and locate the flames in the plant in real time. The detecting frame rate can reach 15 ms/frame, and the detection rate reaches 100%, which has high stability. An efficient and reliable fire monitoring solution is provided for the chemical industry.