In the operation of petrochemical plants, safety operation is the top priority, especially for the control of fire operation. In this paper, based on the real hot work in petrochemical plants, based on the extraction methods of different edge operators, the compound analysis of different edge operators can be carried out to visually detect the hot work of petrochemical workers in petrochemical plants, and based on the neural network algorithm, the convolution operation of edge data is carried out, and the data is input into the hot work database to build the hot work database and identify the flame. Remind the hot operation, according to the size of the flame to make a preliminary judgment on whether the danger, combined with the characteristics of the petrochemical device itself to add smoke identification. This method is of great significance to improve the thermal safety monitoring level of petrochemical plant.