SAR图像ROEWA边缘检测器的改进

发布时间：2018-05-17 03:01

本文选题：合成孔径雷达图像 + 边缘检测　；参考：《遥感学报》2017年02期

【摘要】：针对指数加权均值比(ROEWA)边缘检测算子无法准确确定边缘位置和边缘方向的问题,对其表达式进行了改进。重新定义了反转的、带符号的、归一化最小指数加权均值比(IROEWA),作为边缘强度指标,来量化描述边缘的跃变程度,并在此基础上准确计算出边缘的方位。同时将改进的非极大值抑制算法应用到边缘检测流程中,对IROEWA得到的边缘强度图进行后处理。针对自然SAR图像和仿真SAR图像的实验都表明,改进算法能够得到比ROEWA加分水岭分割的原始算法更好的边缘检测结果。利用接收者操作特征(ROC)曲线,对改进算法进行量化评价,其曲线下面积(AUC)可达0.97570,非常接近理想的检测器;在其ROC曲线的最优点处,检测率可达0.95232,而虚警率仅为0.00214左右。
[Abstract]:Aiming at the problem that the edge detection operator of exponential weighted mean ratio (ROEWA) can not accurately determine the edge position and direction, the expression is improved. The inverse, signed, normalized minimum exponential weighted mean ratio (IROEWAA) is redefined as an index of edge strength to quantitatively describe the jump degree of the edge, and on this basis to accurately calculate the orientation of the edge. At the same time, the improved non-maximum suppression algorithm is applied to the process of edge detection, and the edge intensity map obtained by IROEWA is processed after processing. Experiments on both natural SAR images and simulated SAR images show that the improved algorithm can obtain better edge detection results than the original algorithm of ROEWA plus watershed segmentation. The improved algorithm is evaluated quantitatively by using the receiver operating characteristic (ROC) curve. The area under the curve can reach 0.97570, which is very close to the ideal detector, and the detection rate is 0.95232 and the false alarm rate is only about 0.00214 at the best point of the ROC curve.
【作者单位】：地质灾害防治与地质环境保护国家重点实验室(成都理工大学);地学空间信息技术国土资源部重点实验室;成都理工大学地球科学学院;成都理工大学旅游与城乡规划学院;电子科技大学资源与环境学院;
【基金】：中国地质调查局地质矿产调查评价国家专项工作项目(编号:12120113095400)
【分类号】：TN957.52

【参考文献】