无监督纹理遥感图像分形分割仿真

针对无监督纹理遥感图像分形分割时,仅利用光谱信息,没有充分利用遥感图像中的纹理信息造成过分割或是欠分割的问题,提出结合光谱信息和纹理信息的无监督遥感图像分形分割方法。方法首先借助于双边滤波算法建立双边滤波模型,对无监督遥感图像进行平滑处理,利用边缘检测算子获取无监督遥感图像的纹理梯度和光谱梯度,保留遥感图像的纹理信息。将图像纹理梯度和光谱梯度的内积范数作为K-means算法的距离测度,针对K-means算法易陷入局部寻优问题,将遗传算法与K-means算法相结合实现全局寻优,完成无监督纹理遥感图像分形分割。通过在高分辨率遥感图像上的分割对比实验,证明上述方法能充分利用遥感图像的光谱和纹理信息,分割效果良好。

Fractal Segmentation Simulation of Unsupervised Texture Remote Sensing Image

In fractal segmentation of unsupervised texture remote sensing image,the texture information in remote sensing image is not fully utilized,and thus leading to over-segmentation or under-segmentation.Therefore,a fractal segmentation method for unsupervised remote sensing images based on spectral information and texture information was presented.Firstly,this method built the bilateral filter model by bilateral filter algorithm and smoothed the unsuper-vised remote sensing image.Secondly,our method used the edge detection operator to get the texture gradient and spectral gradient of unsupervised remote sensing image,so as to preserve the texture information from remote sensing image.Then,we took the inner product norms of texture gradient and spectral gradient as the distance measure of K-means algorithm.The K-means algorithm was easy to fall into the local optimization,so we combined genetic algo-rithm with K-means algorithm to realize the global optimization and thus to complete the fractal segmentation of unsu-pervised texture remote sensing image.The simulation for high-resolution remote sensing images shows that the pro-posed method can make full use of the spectral information and texture information of remote sensing image.Thus,the segmentation effect is very good.