一种简化脉冲耦合神经网络的高分辨率农村公路影像分割方法

摘要

针对高分辨率图像中地物信息表现更加精细,增大了噪声对分割农村公路的影响,而一般的分割方法容易产生过分割现象的问题,提出基于简化脉冲耦合神经网络(pulse coupled neural network,PCNN)农村公路分割方法.该方法首先采用最小交叉熵确定其迭代次数,然后用典型的简化PCNN模型对图像进行分割,并在此分割基础上,利用形态学方法,根据斑块面积的大小对农村公路进行最终分割提取.通过利用0.2m高分辨率无人机影像进行试验,并与经典算法区域生长法和Hough变换直线检测方法比较.结果表明,该方法可有效地分割出农村公路,避免了图像过分割的缺点,具有目标边缘光滑、连贯和清晰的特点,用于高分辨率图像中农村公路的分割处理效果优于常规方法.定量评价结果表明,该方法在总体精度、Kappa系数上都有一定的提高.%For the information in high-resolution images is more refined, it can increase the impact of noise on rural roads segmentation, and the general segmentation methods always have phenomena of over-segmentation, the method based on simplified pulse coupled neural network (PCNN) is proposed. Firstly, the number of iterations is determined by the minimum cross entropy.Then, the image is segmented by a typical simplified PCNN model.Based on segmented results, the rural road is extracted according to the size of the patch area using the morphological method.It is found that the method can effectively segment the rural roads and avoid over-segmentation by comparing with the classical algorithm-regional growth method and the Hough transform linear detection method.And this proposed method can make the edge of the target smooth and features of rural roads in high-resolution images clear.The results of the quantitative evaluation show that the method has a certain improvement in overall accuracy and Kappa coefficient.