摘要
针对自然纹理背景,提出一种基于多尺度小波特征融合的人造目标检测方法。在小波变换域,纹理背景和目标区域的多级小波系数具有不同的能量分布,能量特征可作为简单、有效的空间特征来检测目标。由于小波函数具有良好的局域性特点,不同尺度下用它检测出的边缘特征点移位不会超过1个像素。融合边缘特征和能量特征进行人造目标检测,可有效地保证目标边界的定位精度,达到较好的鲁棒性和准确性。实验结果证明,该方法对纹理背景下人造目标面积探测的误差率小于5%,目标探测概率大于94.1%。
Based on multi-scale wavelet transform, a multi -feature fusion approach for automatically detecting man-made objects in areas of natural background is proposed. Due to the energy distribution difference between targets and nature texture background in wavelet multiscale decomposing, the energy signature of sub-bands coefficients can be chosen as a simple but effective characterization of spatial texture. The edge feature is based on multiscale decomposition, which offers accurate and precise edge localization. By combining energy with edge information, the proposed algorithm can obtain robust segmentations. Experimental results prove that less than 5% of ultimate measurement error of the target area and more than 94.1% of detection probability for targets can be obtained by the method.
出处
《光电工程》
EI
CAS
CSCD
北大核心
2005年第11期1-4,共4页
Opto-Electronic Engineering
基金
武器装备预研基金资助
关键词
目标提取
多尺度分析
能量特征
边缘特征
特征融合
Target detection
Multi-scale analysis
Energy signature
Edge feature
Feature fusion