基于Fisher两类判别的机库目标识别

Hangar Target Recognition in Two Methods of Fisher Discrimination

为实现根据波谱特征识别高分可见光遥感影像上的机库目标,通过分析影像上机库与其他常规地物之间波谱特征的区别,利用常规的Fisher两类判别法分类识别,并针对错分像元较多的问题构建逐层剥离法进行改进。结果表明:逐层剥离法可识别出77%以上的机库目标像元,错分像元较常规方法减少85%以上,能有效去除影像上大部分其他地物的干扰,具有更高的识别精度和效率。

To recognize the hangar target in high-resolution visible remote sensing image based on spectral features,the difference of spectral features between hangar and other conventional objects was analyzed,the conventional Fisher discriminant method was used to classify and identify the hangar target,and a layer-by-layer stripping method was constructed to solve the problem of more misclassified pixels.Results show that the layer-by-layer stripping method could recognize more than 77%of the target pixels in the hangar,reduce the misclassification pixels by more than 85%compared with the conventional method,remove effectively the interference of most other objects in the image,and reach higher recognition accuracy and efficiency.