基于显著点提取的红外弱小目标检测

根据人类视觉感知理论,采用bottom-up控制策略的预注意机制和top-down控制策略的注意机制,提出了一种适用于自动目标识别的目标检测算法.该算法首先对输入图像进行非均匀区域分割,根据对象的显著性特点,在已分割好的各个区域提取出显著性点作为潜在目标点,得到潜在目标点集合,之后采用改进的双滑窗算法对这个集合进行更为细致的识别,剔除伪目标,检测出真实目标.实验表明,该算法具有良好的检测效果,预注意机制有效降低了算法运行的时间,改进的双滑窗算法使得检测的鲁棒性更强,对于目标区域带有运动阴影的红外图像以及复杂背景下的红外图像均能进行正确的检测.

Detection of small and dim targets in infrared images based on attention mechanism

According to the theory of human visual perception and using pre - attention mechanisms with the bottom - up control strategy and the attention mechanism with the top - down control strategy, a target detection algorithm which is suitable for target recognition is proposed. Firstly, inhomogeneous region segmentation is used, based on the significant characteristics of the target object, and the algorithm extracts significant points as the potential target in each region. Then, it uses the improved double sliding window algorithm for more detailed identification. Exper- iments show that this method has a good effect on the dim and small target detection in infrared images, the pre - attention mechanism effectively reduces the running time of the algorithm, the double sliding window algorithm can improve the robustness detection in infrared images, it is suitable for the infrared images of the target area with a moving shadow and infrared images against complex backgrounds.