中文摘要：

根据远距离前视红外图像中机场跑道的成像特点，提出了一种基于仿射不变矩的红外序列图像机场目标识别方法．该方法对图像进行多次双阈值分割，搜索分割图像中各潜在目标区域轮廓链，计算其仿射不变矩和目标背景梯度特征，并利用前后帧目标大小的约束关系等先验知识，对各候选目标区域进行识别，最终获取机场目标．试验结果表明：该算法对复杂背景下的远距离前视机场目标的正确识别率优于96％，误识率低于2％．

英文摘要：

The characteristics of the imaged airfield runways in forward looking infrared images were analyzed. Target recognition by affine moment invariants is proposed for the detection of airfields in forward looking infrared （FLIR） image sequences. First the image is segmented with double-gated thresholds iteratively, then the contour chains of the potential target areas are traced in the segmented images, so the affine moment invariants of the target and intensity gradient of the target boundary in the original image should be calculated, combining them with the prior knowledge of target size restriction in the image sequences, the potential target areas are identified. The experimental results show that the correct airfield identification probability of the proposed algorithm in the condition of complicated background from long distance in FLIR images is over 96 %, besides the false probability is less than 2 %.