科学研究

学术交流

首页 >  科学研究 >  学术交流 >  正文
PolSAR Image Classification and Change Detection Based on Deep Learning
作者:    时间:2018-09-10 浏览次数:

  

报告人: 刘芳

工作单位: 南京理工大学

报告时间:9月14日上午8点

报告地点:数学与统计学院一楼报告厅

报告摘要: Polarimetric Synthetic Aperture Radar (PolSAR) image classification and change detection are two hot topics in Remote Sensing (RS), and they are widely used in monitoring the earth covering and the evolution of the earth surface. In this talk, we first present a new type of restricted Boltzmann machine (RBM) for PolSAR data, i.e., Wishart-Bernoulli RBM (WBRBM), which is used to form a deep network named as Wishart DBN (W-DBN). Numerous unlabeled PolSAR pixels are fully used to model PolSAR pixels by W-DBN. In addition, the coherency matrix is used directly to represent a PolSAR pixel without any manual feature extraction, which is simple and time saving. Then, a specific deep model for PolSAR image classification is proposed to improve the efficiency of the classification algorithm, which is named Wishart DSN (W-DSN). It establishes a simple but clever connection between PolSAR image interpretation and deep learning. Next, a novel version of Generative Adversarial Network (GAN) named Task-Oriented GAN is proposed to tackle small sample problem. Besides that, a novel version of Convolutional Neural Network (CNN) is also presented to detect changed areas in multitemporal PolSAR images, which is named Local Restricted Convolutional Neural Network (LRCNN). It not only recognizes different types of changed/unchanged data, but also ensures noise insensitivity without losing details in changed areas. What’s more, for the task of change detection in large-scale PolSAR image, a new method based on Looking-Around-and-Into mode is proposed, which is inspired by the way of visual searching in an electronic map. It looks around firstly to locate the candidate domain (Look Around) and then analyze them in different scales (Look Into). Experiment results show that the proposed methods perform very well in PolSAR image classification and change detection.

报告人简介:
刘芳,博士,南京理工大学讲师。2012年毕业于河南大学数学与统计学院,获信息与计算科学专业学士学位,并保送至西安电子科技大学硕博连读,2018年毕业于西安电子科技大学人工智能学院,获智能信息处理博士学位。主要研究方向包括模式识别、计算机视觉和深度学习等,研究领域主要为遥感图像处理、地物变化检测和场景理解等方面。在IEEE Transactions on Image Processing、IEEE Transactions on Geoscience and Remote Sensing和IEEE Transactions on Neural Networks and Learning Systems等国际著名期刊和会议上发表论文多篇,并为多个国际著名期刊和会议的审稿人。目前主持国家自然科学基金青年科学基金项目一项,参与国家自然科学基金重点项目、国家自然科学基金面上项目和江苏省重点研发计划项目各一项。