摘要
当前,深度学习技术与传统的面向对象技术相结合的分类方法已经较好地应用于高分辨率遥感影像的分类任务当中,但是仍存在如下问题:高分辨遥感影像地物目标复杂,依靠单一数据源进行分割效果不佳;标准的卷积神经网络只能接受固定尺寸大小的输入,分割对象在拉伸变形至固定尺寸的过程中会造成信息的损失。该文首先结合DSM数据进行协同分割,获得更佳的分割结果;然后将空间金字塔池化层(Spatial Pyramid Pooling,SPP)引入卷积神经网络中,构建了一种能接受任意尺寸输入的深度学习面向对象分类模型,从而令分割对象的特征表达更完整,以提高影像分类精度。实验结果表明:引入空间金字塔池化层的高分辨率遥感影像深度学习分类方法,可有效提高影像分类精度,进而得到更加真实可靠的分类结果。
Currently,the classification method based on the combination of deep learning technology with traditional object-oriented technology has been well applied to the classification of high-resolution remote sensing images.However,there are still the following problems:The segmentation relying on a single data source is poor because of the complex ground objects of high-resolution remote sensing images.The standard convolutional neural network can only accept input of fixed size,the segmentation object will cause the loss of information in the process of stretching and deforming to a fixed size.In the study,the better segmentation results are achieved by collaborative segmentation based on DSM data.Then,the spatial pyramid pooling(SPP)are introduced into the convolutional neural network to construct a deep learning object-oriented classification model that can accept input of any size.It can make the feature expression of segmented objects more complete,and improve the accuracy of image classification.The experimental results show that the deep learning classification method based on spatial pyramid pooling can effectively improve the accuracy of image classification,and then,obtain more real and reliable classification results.
作者
苏成林
郎垚
龚秋全
董武钟
Su Chenglin;Lang Yao;Gong Qiuquan;Dong Wuzhong(Sichuan Electric Power Design&Consulting Co.,Ltd.(SEDC);Faculty of Geosciences and Environmental Engineering,Southwest Jiaotong University)
出处
《勘察科学技术》
2021年第4期20-24,共5页
Site Investigation Science and Technology
基金
四川省科技厅重点研发项目(2017SZ0027)
四川电力设计咨询有限责任公司科技项目(kj2021-k-1)
