主講人：Chein-I Chang, Department of Computer Science and Electrical Engineering, University of Maryland, Baltimore County, Baltimore, USA
時間：2021年7月19日9：00-11：00(北京時間)
線下參與方式：機器人視覺感知與控制技術國家工程實驗室 305報告廳集中聽課
線上參與方式：
加入 Zoom 會議鏈接: https://us05web.zoom.us/j/87357345709?pwd=VUc1SUNmL3dqcndnSUlCeFVOZWtVUT09
會議號：873 5734 5709 密碼：123
主講人簡介:
Chein-I Chang 教授于1987年畢業于美國馬里蘭大學帕克分校，電機工程專業博士。現任美國馬里蘭大學巴爾地摩郡分校電機工程系終身教授，同時也是IEEE Life Fellow與SPIE Fellow、大連海事大學講座教授、臺灣中興大學遙測科技杰出講座教授，并擔任IEEE Transaction on Geoscience and Remote Sensing、Remote Sensing等多個國際知名期刊編委，已發表SCI檢索學術論文200余篇，其中超過100篇屬于高光譜領域，撰寫高光譜領域專著4部，并編著高光譜領域書籍3部，授權美國專利7項，Google被引次數達25000多次，Google學術H指數為66。
講座簡介:
Hyperspectral target detection can be performed in two different modes, active detection such as known target detection and passive detection such as anomaly detection. To evaluate detection performance, a general criterion is to use the area under a receiver operating characteristic (ROC) curve, AUC which is plotted based on detection probability, PD versus false alarm probability, PF. Unfortunately, Unfortunately, many ROC curves reported in the literature are indeed incorrectly generated. Another major issue is that using AUC of a ROC curve of (PD,PF), denoted by AUC(D,F) is unreliable and misleading because PD and PF are generated by the same threshold. As a result, a higher PD also generates a higher PF and vice versa. To address these two issues this talk presents a 3D ROC analysis which generates a 3D ROC curve as a function of (PD,PF,?) by including the threshold parameter as a third independent variable. Consequently, a 3D ROC curve along with its derived three 2D ROC curves of (PD,PF), (PD,?) and (PF,?) can be further used to design new quantitative measures to evaluate the effectiveness of a detector and its target detectability TD and background suppressibility (BS). To demonstrate the full utility of 3D ROC analysis in target detection, examples are included in this talk to demonstrate how 3D ROC curves can be used to design new detection measures to evaluate target/anomaly detection performance more effectively ad accurately in terms, TD, BS and detector’s effectiveness.
