Scene Classification in High Resolution Remotely Sensed Images Based on PCANet

摘要：

　　Rich information provided by high resolution remotely sensed images allow us to classify scenes by understanding their spatial and structural patterns.The key of scene classification task with remotely sensed images lies in feature learning efficiency and invariant image representations.While deep neutral network-based approaches achieved good classification accuracy for remotely sensed images,they often have to train millions of parameters and involve heavily iterative computation.In this paper,we propose a new framework for scene classification based on a simple PCANet which is introduced into high remotely sensed image classification for the first time.First,we verify the eligibility of PCANet on classifying large scale scenes from high resolution remotely sensed images.Then we explore the impact of PCANet parameters including filter size,number of filters,and block overlap ratio on classification accuracy.Lastly,we do comprehensive experiments with the public UC-Merced dataset to exemplify the effectiveness of the approach.Experimental results show that the proposed framework achieved on par with the state-of-the-art deep neutral network-based classification accuracy without training a huge amount of parameters.We demonstrate that the proposed classification framework can be highly effective in developing a classification system that can be used to automatically scan large-scale high resolution satellite imagery for classifying scenes.

关键词： Deep learning Image classification High resolution remotely sensed image PCANet Feature learning

作者: Dongmei Huang Yanling Du Qi He Wei Song Kefu Liu

作者单位: College of Information Technology,Shanghai Ocean University,Shanghai,China East Sea Branch of the State Oceanic Administration,Shanghai,China

会议类型: 国际会议

会议名称: International Asia-Pacific Web Conference(第18届国际亚太互联网大会)

会议地点: 苏州

会议语种:英文

页码: 179-190

在线出版日期: 2016-09-23（万方平台首次上网日期，不代表论文的发表时间）