基于机器学习的MRTD客观测试方法研究  

Research on MRTD objective testing method based on machine learning

作　　者：季然 肖茂森[1] 李硕[1,2] 刘宇 罗湛仪[1,3] 程嘉维 JI Ran;XIAO Maosen;LI Shuo;LIU Yu;LUO Zhanyi;CHENG Jiawei(Xi'an Institute of Optics and Precision Mechanics,Chinese Academy of Sciences,Xi'an 710119,China;.2.School of Optoelectronics,University of Chinese Academy of Sciences,Beijing 100049,China;School of Physics and Information Technology,Shaanxi Normal University,Xi'an 710119,China)

机构地区：[1]中国科学院西安光学精密机械研究所,陕西西安710119 [2]中国科学院大学光电学院,北京100049 [3]陕西师范大学物理学与信息技术学院,陕西西安710119

出　　处：《系统工程与电子技术》2024年第10期3265-3270,共6页Systems Engineering and Electronics

摘　　要：红外成像技术的加速发展对红外成像系统测试和评估的客观性和准确性提出了更加严格的要求。针对当前红外成像系统最小可辨温差(minimum resolvable temperature difference,MRTD)存在的测试主观性、操作复杂性等问题,提出基于支持向量机(support vector machine,SVM)和卷积神经网络(convolutional neural network,CNN)的两种MRTD客观测试方法。通过引入数据增强技术,避免由训练样本少及网络层次复杂导致的过拟合。实验结果表明,与实际人员对数据的判断相比较,MRTD测试使用SVM方法的识别准确率为94.50%,训练时间为8.22 s;CNN方法3次训练平均准确率为99.07%,迭代100次训练时间为487.48 s。SVM方法的实时性更好,CNN方法具有准确率高的特点,实验结果验证了这两种MRTD的客观测试方法为红外热成像系统性能指标研究提供了一种可靠的量化和评估工具。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.

关 键 词：最小可辨温差 机器学习 深度学习 支持向量机 卷积神经网络 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]