【作者】 张雪晴；

【导师】 于盛林；

【作者基本信息】 南京航空航天大学 ， 测试计量技术及仪器， 2007， 硕士

【摘要】 飞机防结冰技术近几十年得以迅猛发展。研究表明飞机表面材料和过冷水滴,与机翼表面结冰之间存在一定联系。通过飞机冰风洞实验,模拟了过冷水滴结冰环境,研究了不同材料的干湿性能在飞机防结冰上的效果。实验表明干性材料,过冷水滴撞击后不易残留在机翼表面,减小形成冰层的机率,而湿性材料则与之相反。以前,人们用肉眼直接从过冷水滴飞溅图像中,根据水滴飞溅的不同形状,判断材料的干湿性能。由于这种判断方法效率不高,以及人为判断的不一致性,从而引入数字图像处理方法处理实验图像,来判断材料的干湿性能。本文在实验中采集了大量的过冷水滴飞溅图像,研究了过冷水滴在撞击飞机表面过程中水滴飞溅的现象,定义了一个多维特征空间用于描述图像中飞溅水滴的特征。这个特征空间由图像的平均亮度,均熵,熵的方差,霍夫变换以及相关统计值组成。通过200张试样图像的分析,得出水滴飞溅图像的类别。结合k均值聚类算法将图像分为干表面图像,湿表面图像,干湿混合表面图像三个类别。整个飞机冰风洞实验,共测试了9个标准试件。根据图像分类的结果判断试件材料的干湿性,8个试件能正确地分为干或湿状态,1个试件发生误判。研究结果表明自动分析图像较好地解决了由于手工分析所造成的耗时及准确性问题,试件分类的正确率达到89%。更多还原

【Abstract】 The research on super-cooled large droplets (SLD) has been developing dramatically in recent years. It proves that super-cooled large droplets impingement onto airfoil and aircraft materials have a great effect on aircraft icing. In the research facility of icing wind tunnel at Cranfield University, it aims to classify images into wet/dry surface with the intention of sample identification. The droplets ran quicker onto dry samples so that it could protect airfoil from icing. In the past, droplet splashing images were arranged manually to identify the surface condition, which simulated the aerofoil in SLD icing environment. However, due to the inefficiency of physical analysis and inconsistent perspectives on the same droplet splashing image, digital image processing techniques are used to technically improve image classification in accordance with droplet splashing models.A great number of droplets splashing images are captured through aircraft icing experiments. This thesis defines a multi-dimensional feature space as average brightness, average entropy, standard deviation of entropy and Hough line, with a ratio of entropy and Hough lines to characterize the images as criteria of classification. Based on manual analysis of 200 training images, classification group has been defined as four. After the organization by k-means algorithm, images can be categorized into dry surface, wet surface, and ambiguous images.Among 9 samples, eight samples succeed to be identified into wet/dry behavior. Meanwhile, one sample is to false classification since the raw images are insufficient to represent the entire droplets splash impact events. However, there are three samples left to be validated by humans. The droplet splash images can be processed quickly and correctly, due to time consumption by human classification.更多还原