In order to reduce nitrogen oxides in the earth's atmosphere caused by pollution. Thermal power plants are gradually adding a selective catalytic reduction denitrification technology during the operation of the power plant. But in the power plant in the process of the denitration, the phenomenon of sulfur dioxide (SO2) being oxidized to sulfur trioxide (SO3) occurs and cannot be accurately monitored. Therefore, in order to have a more intuitive understanding of SO3 emissions from power plants, we use a transfer learning based on VGG16 network to study it, which is of significant reference value for the application of machine learning techniques in predicting atmospheric pollution from thermal power plants. Real smoke data is put into the model to make predictions, analyze and verify the effects, and compare the predicted effects with other machine learning models showing that the transfer learning model has higher identification accuracy.