摘要
针对目前农村宅基地传统调查方法存在的工作量大、周期较长等不足,本文提出了一种顾及纹理和形状特征的农村宅基地半自动提取方法。该方法首先利用Gabor小波变换产生纹理特征矩阵,与光谱特征共同训练目标地物特征的混合高斯模型,并改进直线Hough变换算法检测道路范围,可有效区分农村宅基地与道路目标。无人机遥感影像实验证明,本文所提出的提取方法可有效克服传统方法结果中的斑点噪声和误分类现象,大大提高宅基地分类精度,Kappa系数达0.8以上,耗费时间也有明显缩短。
In order to cut the high cost of traditional rural residential area survey, a kind of texture and shape feature re- ferred semi-automatic method,is presented to extract the rural residential area. Firstly, a texture feature matrix is generated by using the Gabor wavelet transform. With the spectral feature, a multi-dimensional feature matrix is constructed according to this algorithm in order to describe the feature of object in the image. Secondly, the region of geo-object is marked in the image. By reading the corresponding vector of the feature matrix, the mixture Gaussian model for the object is then trained. With this trained mixture Gaussian model, rough extraction result could be obtained. Finally, by the result of Hough transform counter clustering,incomplete lines in the image can be made up. The second Hough transform will discriminate between rural roads and rural residential area,and get the accurate result by getting rid of the wrong geo-object. A studied area in Fengdu, Chongqing was taken into experiment. The result showed that this method could effectively remove the impact of spectral similar and salt noise phenomenon, and greatly improve the extraction accuracy of rural residential area. The Kappa coefficient of the experiment result was 0.85 ,and the time cost was reduced by 46~.
出处
《遥感信息》
CSCD
2013年第4期37-43,共7页
Remote Sensing Information
基金
国家科技支撑计划课题(2012BAJ23B05)