报告名称：Integral Probability Metric based Autoencoder for Hyperspectral Unmixing

报告专家：李红

专家所在单位：华中科技大学

报告时间：2021年12月29日下午2：30-4：30

报告地点:学院201报告厅

专家简介：李红，教授，博士生导师，湖北省名师，华中卓越学者特聘教授及享受国务院政府特殊津贴专家。科技部国际科技合作计划评议专家，湖北省计算数学学会理事，美国IEEE会员。主要从事逼近与计算、机器学习与模式识别等方面的研究，在IEEE Trans等重要学术期刊上发表学术论文60余篇。主持国家自然科学基金、“十二五”航天支撑计划项目及国防预研基金等多个科研项目。2006年至2021年期间多次应邀访问香港浸会大学、澳门大学、美国加州大学尔湾分校（UCI）、澳大利亚悉尼大学等，十余次出席国际学术会议。2006年获宝钢教育基金“优秀教师”奖；2009年主持建设的“复变函数与积分变换”课程被评为国家精品课程、2016年评为国家精品资源共享课程、2018年评为国家精品在线开放课程及2020年被评为国家一流课程。

报告摘要：Hyperspectral unmixing is a significant task in the remote sensing image analysis. In this talk , a joint metric neural network (JMnet) is proposed for hyperspectral unmixing, by introducing Wasserstein distance and feature matching as regularization terms, and SAD as the underlying loss. The proposed neural network consists of two parts, an autoencoder is used for endmember extraction and abundance estimation while a discriminator to compute the Wasserstein distance. The Wasserstein distance can stably provide useful gradient information that promotes the autoencoder to reach a solution with better unmixing performance. The feature matching is adopted to an intermediate layer of the discriminator for enforcing the features of the observation and the reconstruction to be equal, which can lead to further improvement of the unmixing performance. Model analysis and regularization parameteranalysis are conducted to demonstrate the effectiveness of ourmethod. Experimental results on four real-world hyperspectraldata sets show that our method outperforms the state-of-the-artmethods, especially in terms of abundance estimation.

（审核：郑大彬）