论文部分内容阅读
本文介绍用于红外前视(FLIR)实时目标识别的特征分析及分类器设计。对低分辩率FLIR背景中候选战车目标进行提取之后,在一次通过时从目标地域中抽出一组17个特征。这些特征包括2个光强度特征,9个几何特征,6个结构特征。本文研究了5种类型的候选目标,它们是坦克、装甲运兵车,吉普、正在燃烧的建筑物和其它比如噪声干扰区一类非目标。[根据人工对特征的描述设出简易树形分类器,即对树形分类器中每一非结节(点)的特征在目标种类中的分布作出描述]。这种特征抽出和分类器的设计方法,已应用在美国陆军微光摄像实验室研制的FLIR成像仪上,并取得预期结果。
This article describes the characterization and classifier design for FLIR real-time object recognition. After extracting candidate tank targets from a low-resolution FLIR background, a set of 17 features are extracted from the target area on a pass. These features include 2 light intensity features, 9 geometric features, and 6 structural features. This article examines five types of candidate targets: tanks, armored personnel carriers, jeeps, burning buildings, and other non-targets such as noise-distracting zones. [Describe the simple tree classifier according to the description of the artificial features, that is, describe the distribution of each non-nodule (point) feature in the target class in the tree classifier]. This feature extraction and classifier design method has been applied to the FLIR imager developed by the Army’s Low Light Imaging Laboratory and achieved the desired results.