Abstract:

The accelerated development of infrared imaging technology has put forward more stringent requirements for the objectivity and accuracy of the testing and evaluation of infrared imaging systems. Aiming at the current problems of test subjectivity and operational complexity of the minimum resolvable temperature difference (MRTD) of infrared imaging systems, two MRTD objective test methods based on support vector machine (SVM) and convolutional neural network (CNN) are proposed. By introducing the data enhancement technique, the overfitting caused by the small training samples and the complex network hierarchy is avoided. The experimental results show that compared with the actual personnel's judgment of the data, the MRTD test using the SVM method has a recognition accuracy of 94.50% and a training time of 8.22 s, while the CNN method has an average accuracy of 99.07% in three training sessions, and a training time of 487.48 s for 100 iterations. The SVM method has better real-time performance and the CNN method is characterized by high accuracy. The experimental result verifies that these two objective test methods of MRTD provide a tool for quantification and evaluation of infrared thermal imaging system performance indicators research.

Key words: minimum resolvable temperature difference (MRTD), machine learning, deep learning, support vector machine (SVM), convolutional neural network (CNN)