Abstract:

The classification of High-Resolution Remote Sensing images（HRRS） is a challenging task. Aiming at the semantic characteristics of remote sensing dataset images, a Convolutional Neural Network（CNN） with random sub-image extraction of the dataset images and a pyramid pooling model is proposed. The size of the input image is cut based on the random size of Cauchy distribution, and these sub-images with different sizes but the same label are sent to a convolutional neural network with Spatial Pyramid Pooling（SPP）. The predicted category mode is output as the final classification. The experimental results show that the proposed method improves the classification accuracy of multi-class remote sensing images.

Key words: remote sensing image, Cauchy distribution, sub-image, convolutional neural network, space pyramid pooling

摘要：

高分辨率遥感图像（HRRS）的分类是一项具有挑战性的任务。针对遥感数据集图像本身的语义特性，提出一种对数据集图像进行随机子图像提取并带有金字塔池化模型的卷积神经网络（Convolutional Neural Network，CNN）。对输入图像的尺寸进行基于柯西分布的随机尺寸剪切，将这些尺寸不同但是标签相同的子图像送进带有SPP（空间金字塔池化）的卷积神经网络，将子图像的预测类别众数作为最终分类输出。实验结果表明该方法对多类遥感图像的分类精度有一定提升。

关键词: 遥感图像, 柯西分布, 子图像, 卷积神经网络, 空间金字塔池化