基于区域显著性与稳定性的小目标检测方法

Small Target Detection Method Based on Regional Stability and Saliency

为解决彩色图像小目标检测中目标易丢失与虚警率高的问题,提出了一种基于区域显著性和稳定性标准增强的小目标检测方法(RSSEM)。首先,在区域稳定性特征提取阶段,针对滤波导致的边缘信息缺失问题,填充图像边界并采用多级阈值二值化图像,在聚类准则下二值图像进行区域聚类和二次后验,使本文方法对小目标有较高敏感度。其次,在区域显著性特征提取阶段,利用旋转对称高斯高通滤波对灰度图像进行滤波得到显著性特征图像。最后,融合稳定性特征与显著性特征,并对强噪声滤波后实现小目标检测。在RSS数据集上,与对照组相比,本文方法能显著降低小目标的丢失率和虚警率,比最先进的算法在精确度、召回率、F值上至少提高1%,表明RSSEM的有效性。

To solve the problems of target missing and high false alarm rate in the detection of small targets from color images,we proposed a small target detection method on the basis of regional saliency and stability enhancement metrics(RSSEM).First,aimed at the missing edge information caused by filter,the image boundaries were filled and image binarization was performed by multi-level thresholds at the stage of regional stability feature extraction.The detection of small target was improved as region clustering and secondary posterior were performed on the binary image.Second,the grayscale image was filtered with the rotational symmetric Gaussian high-pass filter to obtain the salient feature image at the stage of regional saliency feature extraction.Finally,the stability features and saliency features were merged,and small targets were detected after filtering strong noise.On the regional saliency and stability dataset,the rates of target missing and false alarm decrease significantly compared with the control group.The proposed method is at least 1%higher in accuracy,recall and F score in comparison with state-of-the-art methods,which indicates the effectiveness of RSSEM.