应用基于纹理的面向对象分类模糊方法提取水田信息

针对面向对象分类方法进行图像处理时出现大量冗余多边形及其分类结果出现斑点等不足,提出基于纹理特征的分割方法。先对原始图像进行纹理增强、主成分分析、低通滤波和图像分割等预处理,得到原始图像的纹理特征二值图;将二值图作为独立波段与其他波段进行假彩色合成,得到纹理增强图,再对其进行多尺度分割和模糊分类。以南京市浦口区2002年9月SPOT图像为例,分别利用上述两种方法提取图像中的水田信息,基于纹理特征的分割方法克服了面向对象分类方法的不足,并且由于在进行模糊分类时分割出的对象大量减少,一定程度上减小了确定模糊区间的难度,较好地提取出水田信息,提高了图像的分类精度。

Extraction of Paddy Field Using Objects-Oriented Fuzzy Classification Approach Based on Texture Characteristic

Redundancy polygons and spots appeared in the classification result when using object-oriented approach to classify remote sensing images.To overcome the shortcoming,a texture-based segmentation approach was proposed.Firstly,pretreatment to original image was taken to obtain texture character image,which includes texture enhancement,principal component analysis,low pass and slice segmentation and so on.Secondly,taking the character image as single band to combine with other bands and obtain texture-enhanced image,and then carry out multi-scale segmentation and fuzzy classification.Two approaches were tried to extract the paddy field of Pukou district in Nanjing with SPOT image in September 2002.The result indicates that:compared to the original approach,the proposed approach improves the classification quality.It avoids the phenomenon of massive redundant polygons and many spots appearing in the classification image.As a result of fewer polygons,the difficulty of determining an accurate fuzzy interval is reduced and the classifying precision increased to a great certain.