模糊神经网络高分辨率遥感影像监督分类

发布时间：2018-03-01 04:39

本文关键词： 高分辨率遥感影像分类模糊神经网络高斯隶属函数监督学习直方图拟合　出处：《中国图象图形学报》2017年08期 　论文类型：期刊论文

【摘要】：目的针对高分辨率带来的像素类属不确定性增大及各类属间相关性增强引起的影像分类问题,提出一种模糊神经网络高分辨遥感影像监督分类方法。方法提出的模型为包含输入层,隐含层(隶属函数层)及输出层的三层前向模糊神经网络,输入层用于接收来自训练样本的灰度值;隐含层每个神经元节点的模糊隶属函数为对各类别定义的高斯隶属函数模型,以实现对输入变量隶属程度的不确定表达;输出层的输入变量为隐含层各神经元节点输出变量的线性组合,激活函数为分段线性函数,该层实现输入变量隶属程度的相关性表达。以训练数据直方图作为期望输出,梯度下降法求解模型参数,最后按最大隶属度准则实现分类决策。结果利用本文算法和经典算法对合成影像进行实验,本文方法总体精度达到0.931,相对于高斯隶属函数方法总体精度提高了5.3%,相对于最大似然法提高了4.2%,相对于FCM方法提高了5.9%,对真实World View-2全色影像的实验中文中方法分割精度也高于传统方法。结论提出的模糊神经网络模型可以更加精确的拟合高分辨率遥感影像复杂的分布特征,有效处理高分辨率遥感影像的上述分类问题。
[Abstract]:Objective to bring high resolution pixels is increased uncertainty and various species classification problem caused by the correlation enhanced image, proposed a fuzzy neural network for high resolution remote sensing image classification method. The proposed model for the method includes input layer, hidden layer (membership layer) and output layer three layer feedforward neural fuzzy the network input layer, for receiving the training samples from the gray value; implicit fuzzy membership function of each layer neuron node membership function model of each category is defined as Gauss, in order to achieve the input variable membership degree of uncertainty expression; input variables of the output layer as a linear combination of the hidden layer neurons of output variables, activation function as a piecewise linear function, the correlation between the expression of layer input variables. The membership degree of the training data as the desired output histogram, gradient descent method is used to solve the model Type parameters, and finally achieve the classification decision according to the maximum membership criterion. The results using this algorithm and the classical algorithm experiments on synthetic images, the overall accuracy reached 0.931, compared to the Gauss membership function method improves the overall accuracy by 5.3%, compared with the maximum likelihood method is increased by 4.2%, compared with the FCM method to improve the method of experiment 5.9%, the real Chinese World View-2 panchromatic image segmentation accuracy is higher than that of the traditional method. The fuzzy neural network model proposed can conclusion distribution more accurately fitting the complex high resolution remote sensing images, the classification of the effective processing of high resolution remote sensing image.

【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,LJYL012) 教育部高等学校博士学科点专项科研基金项目(20122121110007)
【分类号】：TP183;TP751

【相似文献】