The FeO content of sinter is a comprehensive index of sinter production with many factors affecting it and a nonlinear relationship between the factors, which makes it difficult to predict the FeO content. In order to solve the problem that the FeO content of sinter is difficult to predict, a system based on machine vision technology to realize the online perception of FeO content in sinter is proposed. The system obtains the thermography picture information of the tail section by installing infrared thermography equipment at the tail of the sintering trolley, extracts and analyzes the picture information features, and uses the extracted image features of the tail section as the input parameters of the Darknet-19 algorithm, and establishes a prediction model of sinter FeO content based on the improved Darknet-19 algorithm to realize the real-time prediction of sinter FeO content. The field application shows that when the deviation between the predicted value and the actual value in the sinter FeO content prediction model is ±0.5, the accuracy rate is 82.5%, which has a positive effect on stabilizing and optimizing the control of the sintering production process.