红外遥感图像舰船尾迹提取及检测

中低空间分辨率的红外遥感图像中，海面上的舰船所占像素很少，目标的几何形状和具体的纹理结构等特征难以获取。为提升探测的极限信杂比，将呈现线状特征的舰船尾迹特征作为检测要素，对其进行数学表征。创新性地建立了Dot-Curve检测体系，以二维曲率滤波为基础进行初步的舰船检测和尾迹特征提取，建立特征集，从中选取与背景干扰项差异较大的包含尾迹灰度方差、尾迹两侧灰度斜率正负性、尾迹线性度、与船体检测结果距离在内的多项特征，鉴别候选目标的检测结果，去除干扰项，提取目标。结果表明，经过目标鉴别，在不同波段红外图像中，舰船误检率降低到8.40%以下，检测率得到提升，至少达到94.53%。该船只检测算法结合尾迹物理特性和图像特征，适用于多种场景和波段，算法精炼有效，物理规律清晰，所需样本少。

Ship wake extraction and detection from infrared remote sensing images

In infrared remote sensing images with low or medium spatial resolution, the number of pixels occupied by ships on the sea is very small, and the geometric shape and specific texture structure of the target are difficult to obtain. In order to improve the detection limit signal to clutter ratio, the ship wake feature with linear feature was taken as the detection element, which was mathematically characterized. The Dot-Curve detection system was established innovatively. Based on the two-dimensional curvature filtering, the ship detection and wake feature extraction were carried out preliminarily. The feature set was established, from which a number of features with large difference from the background interference items, including wake gray variance, positive and negative gray slope on both sides of the wake, wake linearity and the distance from the hull detection results, were selected to identify the detection results of the candidate targets, remove interference items and extract targets. The results show that after target identification, the ship false detection rate in different bands of infrared images is reduced to less than 8.40%, and the detection rate is improved to at least 94.53%. The ship detection algorithm combines the physical and image characteristics of the wake, which is suitable for many scenes and bands. The algorithm is refined and effective, the physical laws are clear, and the samples needed are few.