利用概率增进树和路径形态学的遥感道路条带提取

Extraction of Road Strip from Remote Sensing Images Based on Probabilistic Boosting Tree Algorithm and Probabilistic Path Morphology

针对高分辨率遥感影像复杂道路提取难题,提出一种利用概率增进树和路径形态学的遥感道路条带提取方法。通过一维Gabor滤波器提取道路角度纹理特征,融合光谱特征构建特征矢量。设计训练样本数据集,利用概率增进树算法提取道路候选点。针对具有一定曲率的复杂道路,兼顾直线和弯曲道路,设计4个主方向邻接图检测线状或条带状道路,改进二值路径形态学为概率路径形态学剔除大多数非道路点。针对小面积噪声和条带孔洞问题,采用数学形态学的方法弥补条带孔洞,得到完整道路条带。结果表明:提取道路条带的准确率达到了88.99%,提取结果较为理想。

The strategy to extract road strip from acquired road stripe image was explored.The workflow is as follows：feature vector is designed by the road angle texture feature which extracted by one-dimensional Gabor filter and the spectral feature.The training sample data set is designed extract the road candidate point by robabilistic Boosting Tree Algorithm.Specifically,for the curved road with a certain curvature,linear or striped roads is detected by 4 main direction adjacency map.Probabilistic path morphology is improved from binary path morphology to remove most non-road points;mathematical morphology is used to make up strip holes,get a complete road strip.The results show that the accuracy of the proposed method is 88.99%,and the result is ideal.