针对传统红外目标检测与识别方法所存在 的问题,即其处理过程总是盲目地对全图进行耗时搜索,提出了一 种基于视觉注意机制和水平集方法的红外海面目标检测与识别方 法。首先,搜索原始图像中的显著性区域,并以获胜点的形式 表示它们。接着,基于所得到的显著性区域,自动初始化水平集函 数,并使演化过程朝着期望的目标轮廓方向挺进,直至演化过程到 达最终的平衡状态。最后,针对远距离(近距离)成像时的输入数 据,给出检测结果(基于不变矩和神经网络框架的识别结果)。对 真实红外海面目标进行的实验证实了本文方法的有效性。

To solve the problem that traditional infrared target detection and recognition methods always blindly carry out time-consuming searching for a whole image in their processing processes, an infrared sea surface target detection and recognition method based on visual attention mechanism and level set method is proposed. Firstly, the salient regions in an original image are searched and are denoted in terms of winner points. Then, on the basis of the salient regions captured, the level set function is initialized automatically and the evolution process is forced to go forward towards the expected target contour until its final equilibrium. Finally, for the input data obtained at a long distance (or at a short distance), the detection results are given (the recognition results based on moment invariants and neutral network framework). The validity of the proposed method is verified in the experiment made on the real infrared sea surface targets.