受到云层飘移、树木晃动、背景噪声杂波等因素的影响，红外弱小目标在地空背景下无法被精确检测。针对这个问题，本文提出了一种融合时空结构张量的背景差分检测算法。首先，通过当前图像与背景模型的比对确定出运动变化的全部像素，再用前景点计数的方法消除噪声等因素造成的孤立点错误检测；然后，时空结构张量模块利用连续帧图像的时间空间信息检测出运动块；最后，对前景目标像素和前景目标块进行融合操作，并将目标区别于背景二值化显示。与其它算法对比的实验结果表明，本文提出的算法具有稳定的检测率，且虚警率有明显降低，是地空背景下红外弱小运动目标检测的有效方法。

Small, infrared moving targets cannot be detected accurately under a ground-sky background owing to the effect of drifting clouds, sloshing trees, and other background clutter. To solve this problem, this paper proposes a background subtraction algorithm integrating the temporal–spatial structure tensor. First, all the changing pixels are distinguished by comparing the current image to the background model. Subsequently, the method of counting the front spots is used to eliminate part of the error-isolated point detection, which is caused by noise or other factors. Further, a structure tensor module uses temporal and spatial information to detect motion blocks. Finally, the foreground target pixels and the foreground target blocks are combined, and the final result is binarized to differentiate the targets from the background. Compared with other algorithms, the proposed one has a stable detection rate and a significant decrease in the false alarm rate. The experimental results show that it is an effective and robust method for small, infrared moving-target detection under a ground-sky background.