基于区间二型模糊神经网络的高分辨率遥感影像分割方法

发布时间：2018-05-04 00:14

本文选题：区间二型模糊模型 + 高分辨率　；参考：《信号处理》2017年05期

【摘要】：高分辨率遥感影像同质区域地物目标异质性增大,光谱测度空间复杂性增加使像素类属的不确定性以及分割决策不确定性增大,引起分割精度下降。提出一种基于区间二型模糊神经网络的高分辨率遥感影像监督分割方法。对同质区域构建一型高斯隶属函数模型刻画像素类属的不确定性;模糊化高斯隶属函数参数构建区间二型模糊模型处理分割决策的不确定性;以训练样本在所有类别中的一型模糊隶属度及上、下隶属度为输入,建立模糊神经网络模型并融入像素邻域关系作为模糊决策。采用文中算法、FCM方法、HMRF-FCM及区间二型模糊神经网络方法分别对合成影像及真实高分辨遥感影像进行分割,定性与定量的对比分析验证了文中算法具有更高的分割精度。
[Abstract]:The heterogeneity of ground objects in homogeneous region of high-resolution remote sensing images is increased, and the spatial complexity of spectral measurement increases the uncertainty of pixel genus and segmentation decision, which results in the decrease of segmentation accuracy. A supervised segmentation method for high resolution remote sensing images based on interval 2 fuzzy neural network is proposed. The uncertainty of pixel genus is described by constructing a type Gao Si membership function model for homogeneous region, and the uncertainty of segmentation decision is dealt with by constructing interval type 2 fuzzy model of fuzzy Gao Si membership function parameter. The fuzzy neural network model is established with the input of the first type of fuzzy membership degree and the upper and lower membership degree of the training sample in all categories. The fuzzy neural network model is incorporated into the pixel neighborhood relationship as a fuzzy decision. The method of HMRF-FCM and interval type 2 fuzzy neural network are used to segment synthetic images and real high-resolution remote sensing images respectively. The qualitative and quantitative analysis show that the proposed algorithm has higher segmentation accuracy.
【作者单位】：辽宁工程技术大学矿业技术学院;辽宁工程技术大学测绘与地理科学学院;辽宁工程技术大学电气与控制工程学院;
【基金】：辽宁省教育厅一般项目(LJYL036) 教育部高等学校博士学科点专项科研基金资助(20122121110007)
【分类号】：TP183;TP751

【相似文献】