Biography
Prof. Guofan Shao
Prof. Guofan Shao
Department of Forestry and Natural Resources, Purdue University, USA
Title: The Right Tool for the Right Job: Map Accuracy vs Classification Efficacy
Abstract: 
Image classification is one of the most challenging tasks in remote sensing applications. Its direct outcome is a thematic map. The quality of thematic maps matters because they are parts of decision making in downstream applications. Researchers have made incredible efforts in continuously advancing image classification techniques to increase the accuracy of classification outcome. However, existing accuracy measures have various problems and can mislead the evaluation of classifier’s performance when imbalanced datasets are involved. There are two common misperceptions: accuracy rates are assumed to comparable across maps and they reflect the discriminative power of classifiers. This is particularly the case for classification with deep learning (DL) because imbalanced datasets tend to undermine the performance of DL models but favor some accuracy measures. Here we discuss and demonstrate the use of image classification efficacy (ICE) to strengthen the evaluation of image classification using DL in remote sensing. The introduction of ICE helps clarify the differences between map accuracy assessment and classifier performance evaluation. Such a differentiation is an important step toward improved research on classifier’s evaluation and advancement.
Biography: 
Prof. Guofan Shao is affiliated with the Department of Forestry and Natural Resources, Purdue University. He serves as Editor-in-Chief for the International Journal of Sustainable Development and World Ecology, published by Taylor & Francis. He started research on geospatial applications in ecosystem modeling and monitoring in 1991 when he became a post-doc at the Department of Environmental Sciences, University of Virginia. Since he joined Purdue University in 1997, he has been teaching and studying remote sensing applications in land use and land cover mapping and its accuracy assessment. He has contributed to research on the mixed forests in eastern Eurasia and central hardwood forests in the US. His current research interest focuses on machine learning classification of remotely sensed imagery acquired from different platforms. He created image classification efficacy as a transformed metric to consistently evaluate the performance of classifiers that deal with different datasets. He has authored or co-authored 6 books, 23 book chapters, and 172 peer-reviewed journal papers. The book chapters include Satellite Data and Remote Sensing in Encyclopedia of Environmetrics in 2012 and 2016, and Optical Remote Sensing in International Encyclopedia of Geography in 2014 and 2019.