基于深度卷积神经网络的高分辨率遥感影像场景分类

场景分类对于高分辨率遥感影像的理解和信息提取具有重要意义。传统方法利用低、中级或语义特征来对影像的场景进行判别，但是针对高分影像涵盖的细节多、类别复杂等特点，中低层特征无法对影像语义进行准确描述。本文提出了一种基于深度卷积神经网络DCNN场景分类模型。首先利用卷积层对影像的纹理、颜色等低阶特征进行提取，然后利用池化层对重要特征进行筛选，最后将提取到的特征进行组合，形成高阶语义特征，利用高阶语义特征对高分影像进行场景分类。为了解决模型的过拟合问题，使用了数据增广、正则化及Dropout提高模型的泛化能力。本文方法在UC Merced-21取得了91.33%的准确率，相比于传统方法，有效地提高了分类精度，同时证明了深度卷积神经网络在遥感影像分类领域优越性。

Scene classification of high-resolution remote sensing image based on deep convolution neural network

Scene classification makes great sense to the understanding and information extraction of high-resolution remote sensing images. The traditional method has used low-level, middle-level or semantic features to distinguish the class of the image scene, but the low or middle level features can't exactly describe the image which are more detailed and complex. In this paper, a DCNN scene classification model based on deep convolution neural network is proposed. The convolution layer is used to extract the image texture, color and other low-level features firstly. Then we use the pool layer to select important features. Finally, the extracted features are merged into high-level semantic features which are used to classify the high resolution remote sensing images. To solve the problem of over fitting, data augmentation, regularization and Dropout are used to improve the generalization ability. This method has obtained 91.33% accuracies on UC Merced-21. Compared with traditional method, the classification accuracies is effectively improved. At the same time, the superiority of deep convolution neural network in remote sensing image classification is proved.