摘要：
In order to improve the accuracy of road information change detection,a new road information change detection method based on fractional integral and spatial neighborhood fuzzy C-means (FCM) algorithm was presented.Firstly,a new difference image was generated by the gray difference calculation of the dual phase remote sensing images after registration and geometric correction.Then,a smaller fractional integral order was used to construct the denoising image mask with eight directions on the upper and lower,left and right,and four diagonals,and the fractional integral calculation were applied to the difference images,which improved the image signal-to-noise ratio (SNR) while preserving the edge and texture details of the image.Finally,the FCM clustering method combined with neighborhood spatial information was used to calculate the difference image after denoising.The highest and lowest points of the difference image gray values were selected as the center point of cluster initialization.The Euclidean Metric of the neighborhood were used to depict different weight values,so as to characterize the influence degree of domain pixels on central pixels and eliminate invalid isolated points.Detecting probability,false alarm rate and missed alarm rate of the algorithm were evaluated by the experiment.The results show that FCM road information change detection method based on fractional integral and neighborhood spatial information can effectively extract road change information.When the integral fractional order is 0.2,the FCM smoothing parameter is 2.5,the detection probability is higher than the comparison algorithm by 18％ to 46％,the false alarm rate is lower than the comparison algorithm by 15％ to 38％,and the missed alarm rate is lower than the comparison algorithm by 3％ to 7％.The present algorithm can achieve better results in suppressing noise information and enhancing texture details.Especially,when the center pixel is noise,due to the introduction of neighborhood information,and it is affected by the neighborhood normal pixels.The proposed method could avoid misclassification by adjusting the membership automatically,it can effectively suppress the influence of neighborhood noise points on the normal pixel classification,and reduce the false alarm rate.2 tabs,4 figs,28 refs.%为了提高道路信息变化检测的精度,提出了一种基于分数阶积分和空间邻域信息的模糊C均值(fuzzy C-means,FCM)聚类算法,用于道路特征变化检测.首先,对经过配准和几何校正的双时相遥感图像进行灰度差值运算,重新生成差值图像;其次,采用较小的分数阶积分阶次构造上下、左右、4个对角线8个各向同性的去噪图像掩模,对差值图像进行分数阶积分计算,在提高图像信噪比的同时保留图像的边缘和纹理细节信息;最后,针对去噪后的差值图像采用结合邻域空间信息FCM聚类进行计算,聚类初始化中心点选择差值图像灰度值最高和最低的点,利用邻域的欧式距离来刻画不同的权重值,以表征领域像素点对中心像素的影响程度,剔除无效孤立点.通过检测概率、虚警率、漏警率指标对提出的算法进行评价.结果表明:基于分数阶积分和邻域空间信息的FCM聚类道路信息变化检测方法能有效提取道路变化信息,当积分分数阶数取0.2,FCM平滑参数取2.5时,检测概率高出对比算法18％～46％,虚警率指标低于对比算法15％～38％,漏警率低于对比算法3％～7％;提出的算法在抑制噪声信息和增强纹理细节方面能够取得较好的效果,尤其当中心像素点为噪声时,由于引入邻域空间信息,其受周围邻域正常像素点的影响,能够自动调整隶属度,较好地避免误分类;该方法还能有效抑制邻域噪声点对正常像素分类造成的影响,降低虚警率.
