摘要
提出了一种基于改进的阶次规正不变矩卫星图像小目标识别方法。首先将卫星图像分割成子块,以图像灰度方差描述子块图像特征,应用所提出的子块合并理论进行分类,减少了卫星图像识别的计算量,大大降低了误判率。提出了改进的阶次规正不变矩理论,并将其应用于小目标物体识别中。以改进的阶次规正不变矩特征作为检测模板和待识别小目标图像相似度的测度,有效区分了小目标物体间的较小差别并解决了由噪声所造成的不封闭性问题;同时将GA理论引入图像匹配识别中。实验结果表明:所提方法识别率可达96.67%,该方法的提出对于图像自动识别具有非常重要的现实意义。
A recognition method of small target based on improved order - normalized moment invariants is improved, which separates the satellite image into subimages, and greyness variances are used to represent subimages feature. Meanwhile, the subimage patterns are classified by applying the theory of subimage consolidation rapidly and efficiently. A recognition theory based on improved order - normalized moment variants is improved. The theory, taking the improved variances for the similarity metrics, distinguishes the unconspicuous difference efficiently, and unclosed structures caused by noises are solved. GA is applied to the recognition. Experiment results show that the accurate recognition rate is 96.67%. This method is of great practical significance.
出处
《计算机仿真》
CSCD
2006年第1期168-171,245,共5页
Computer Simulation
关键词
图像分割
不变矩
阶次规IT
遗传算法
模式识别
Image segmentation
Moment variants
Order- normalization
GA
Pattern recognition