论文部分内容阅读
针对传统浅层机器学习方法应用于高分辨影像分类时存在的问题,提出了结合最小噪声分离变换和卷积神经网络的高分辨率影像分类方法。采用最小噪声分离分析非监督训练初始化卷积神经网络,为提高训练速度,使用线性修正函数作为神经网络的激活函数;利用概率最大化采样原则减少池化过程中影像特征的缺失,并将下采样后影像特征输入Softmax分类器进行分类。采用所提分类方法对典型地区的影像进行分类实验,并与支持向量机和人工神经网络分类方法的分类结果进行对比。结果表明,所提分类方法的分类精度明显高于另两种分类方法的分类精度,并能充分挖掘高分辨遥感影像的空间信息。
Aiming at the existing problems of the traditional shallow machine learning method in high resolution image classification, a high resolution image classification method combined with minimum noise separation transform and convolution neural network is proposed. In order to improve the speed of training, the linear correction function is used as the activation function of neural network. The principle of maximizing the probability of sampling is used to reduce the lack of image features in the process of pooling, and the downsampling Post-image features are entered into the Softmax classifier for classification. The proposed classification method was used to classify the images of typical regions and compared with the classification results of SVM and artificial neural network classification methods. The results show that the classification accuracy of the proposed classification method is obviously higher than that of the other two classification methods, and the spatial information of high resolution remote sensing images can be fully tapped.