改进神经网络的红外成像测温算法  

Algorithm of measuring temperature with infrared imager based on improved neural network

作　　者：李云红[1] 孙晓刚[2] 王延年[1] 张龙[1] 

机构地区：[1]西安工程大学电子信息学院,陕西西安710048 [2]哈尔滨工业大学电气工程及自动化学院,黑龙江哈尔滨150001

出　　处：《红外与激光工程》2010年第5期801-805,共5页Infrared and Laser Engineering

基　　金：陕西省教育厅自然科学专项(10JK571);中国纺织工业协会科技指导性项目(2008012);陕西省自然科学基础研究基金(2009JQ8022);西安工程大学2009年度校基础研究基金项目资助课题(09XG06)

摘　　要：根据热辐射理论和红外成像仪的测温原理,推导了被测物体表面真实温度的通用计算公式以及红外图像的热值与真实温度的对应关系。针对神经网络温度计算法在计算温度时存在较大误差的问题,提出用最小二乘法和改进输入的神经网络法计算温度。这两种算法以红外成像仪输出的三基色之间的比值为自变量或输入量,更好地体现了比色测温原理,因此能较好地消除在实际测温过程中发射率、烟尘和火焰脉动对计算结果带来的影响。通过仿真结果表明,这两种方法的计算精度均高于原神经网络法的精度,而改进输入的神经网络温度算法的计算精度略高于最小二乘法的计算精度。According to the principles of thermal radiation and temperature measurement with infrared imager,a general computing formula was deduced for the measurement of surface temperature and the corresponding relationship between the thermal value and the true temperature of infrared images was investigated.A least squares method（LSM） and an improved neural-network method was developed to calculate the temperature to diminish the deviation of neural-network method.Both the two methods used the ratios among the three basic colors output from the infrared imager as the independent variable or input variable which can personalize the colorimetric temperature-measurement algorithm to reduce the deviation from emissivity,soot and combustion flame on the temperature result.Simulation results show that the precision of these two methods are higher than that of the traditional neural-network method.In addition,the precision of the proposed neural-network method is higher than that of the least squares method.

关 键 词：红外成像仪 表面温度 温度测量 数字图像处理 

分 类 号：TN216[电子电信—物理电子学]