Sinter blanking temperature is an important parameter for judging the sintering endpoint, evaluating the quality of sinter, and affecting the recovery efficiency of sintering waste heat. In order to solve such problems as sinter blanking temperature detection limited by the on-site environment, difficult temperature measurement, instability and poor accuracy, the environmental conditions of the sintering site are simulated by experiments, the test image data are extracted by infrared thermal imaging equipment as the training sample, the thermocouples are used for temperature calibration as the test sample, a non-contact sinter blanking temperature feature extraction method is proposed based on image recognition, and the direct fitting model and the step-by-step fitting model respectively are established. The results show that compared with the data in the test group, the error of the direct fitting model of sinter temperature based on image features is 3.51% and the root mean square error is 28.24, and the error of the step-by-step fitting model of fitting radiant energy and then fitting sinter blanking temperature by luminosity value is 2.10%, and the root mean square error is 17.21. The step-by-step fitting model based on image recognition is more stable and accurate, which has important application value for improving the accuracy and stability of sinter blanking temperature detection, and has a potential role in optimizing the sintering process control and improving the waste heat recovery efficiency.