近岸红外目标检测由于背景复杂和实时性的要求,是当前研究的一个难点。基于上下文的视觉注意机制是解决这一难题的有效途径。首先根据近岸红外目标存在于海面这一先验知识,利用灰度、纹理特征及位置关系,分割出海面;其次利用基于相位谱的视觉注意机制定位显著区域;最后综合海面和显著区域位置,便可检测出近岸红外目标。实验证明,这种算法高效、准确,能够满足目标检测阶段的实时性要求。

Inshore infrared target detection is a difficult problem for the complex background and timesensitivity. Visual attention based on context is effective method for this problem. First, based on the fact that the inshore infrared target must locate at sea, find out the sea by the use of the gray, texture and location characteristics of inshore infrared image. Sencond, locate saliency region by the use of phase spectrum. At last, synthetize the location of sea and saliency region and locate inshore infrared target. Our examples prove this algorithm is feasible and efficient and can meet the requirements of timesensitivity.