论文部分内容阅读
最小可辨温差(MRTD)是评价红外热像仪的重要技术指标,传统的测量方法都是基于测量人员的主观判读得到的,结果重复性差。为了解决这一问题,提出基于人工神经网络的客观评测方法,提出两种特征量——杆均值与背景极值对比度和相邻极值差与均值对比度,使用数字CCD摄像机代替人眼,采集大量图像数据输入人工神经网络,使用BP网络的LM算法进行训练。训练好的神经网络具有类似人的视觉系统的判读能力,可以由计算机代替人工对不同频率、不同温差下的4杆靶图像进行判读。通过Matlab模型和软件验证了系统的可用性,大量测试结果表明:测量结果准确,具有很好的重复性,与相同测量环境条件下由人眼主观测量的MRTD结果相吻合。采用该技术研制的测试设备已成功地应用于某试验单位的检测中,大大提高了测量效率和测量结果的准确性。
The minimum temperature difference (MRTD) is an important technical index for evaluating the thermal imaging camera. The traditional measurement methods are all based on the subjective interpretation of the surveyors, resulting in poor repeatability. In order to solve this problem, an objective evaluation method based on artificial neural network is proposed, and two kinds of feature quantities are proposed, that is, the contrast between pole mean value and background extreme value and the difference between adjacent extreme values and mean value. Using digital CCD camera instead of human eye, The image data is input to the artificial neural network and trained using the LM algorithm of the BP network. The trained neural network has human-like visual system interpretation ability, which can be manually replaced by computers to interpret the 4-bar target images at different frequencies and different temperature differences. The availability of the system was verified by Matlab model and software. The results of a large number of tests show that the measurement results are accurate and reproducible, which is consistent with the MRTD results subjectively measured by the human eye under the same measuring conditions. The test equipment developed by this technology has been successfully applied to the testing of a test unit, which greatly improves the measurement efficiency and the accuracy of measurement results.