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

Interval Type-2 Fuzzy Based Neural Network for High Resolution Remote Sensing Image Segmentation

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。

This paper presented a supervised image segmentation approach based on interval type-2 fuzzy neural network to overcome the problems brought by high resolution remote sensing images. The interval type-2 model was obtained through blurring the mean and variance of the Gaussian model which characterized the uncertainly of the membership of pixels. Then the fuzzy membership and its upper and lower fuzzy membership of the training samples were used as the input of the neuron network in which the influence of neighbor pixels were taken into consideration to construct a decision model to real- ize the segmentation. The proposed algorithm, FCM and HMRF-FCM algorithm and an interval type-2 fuzzy neuron net- works without spatial relationships were performed on synthetic and real high resolution remote sensing images. The qualita- tive and quantitative analyses demonstrate the efficient of the proposed algorithm, especially for homogeneous regions which contains a great difference in its gray level.