A professor from the University of Houston's Cullen College of Engineering has edited his second book on machine learning and signal processing, identifying it as a “rapidly developing” subject area that interests him immensely.
Dr. Saurabh Prasad, an associate professor of Electrical and Computer Engineering, said he was presented with the opportunity to edit Hyperspectral Image Analysis: Advances in Machine Learning and Signal Processing after giving a tutorial lecture on the topic at the flagship Institute of Electrical and Electronics Engineers conference on signal processing (IEEE-ICASSP) in 2017.
“Based on that lecture, I was invited by the area editors in the computer vision and pattern recognition track at Springer to work on a book on this topic,” he said. “There was a compelling need for this book. The area of machine learning and signal processing is a rapidly developing area, particularly with applications to multi-channel imaging, and hence, the book was geared toward researchers and graduate students, to provide them with a comprehensive resource on emerging developments in this area.”
Dr. Jocelyn Chanussot, a professor in the Signal and Images Department at Grenoble Institute of Technology in France, serves as a co-editor for the book. Prasad said this is the second book he has edited, with the first, Optical Remote Sensing, coming out in 2011. He has also co-authored more than 100 research articles in the subject area.
“This book consists of 15 chapters and includes contributions from leading researchers from around the world, including UCLA, Duke, the Los Alamos National Lab, Amazon, the University of Florida and Universitat de València in Spain,” he said.
According to Prasad, topics covered include advances in deep learning, like the design of deep neural networks that characterize the spatial and spectroscopic properties of the imagery data; pixel unmixing, where researchers address and understand composition of mixed pixels that arise when the spatial resolution of the image is poor; data fusion, which is the combination of multiple imaging modalities; limited and noisy ground truth; and other topics.
Asked to describe his work, Prasad said it focuses on advancing and developing machine learning and signal processing techniques to address challenges posed by posed by cutting-edge spectroscopic optical imagers.
“As an example, hyperspectral imaging entails acquiring images that have hundreds to thousands of spectral bands – colors – instead of just the 'Red, green and blue' color channels we are accustomed to with consumer cameras,” he said. “This enables us to see beyond what our eyes can see, and it can very accurately characterize the chemical and biochemical properties of the objects that are being imaged.”
With the “optical signatures” that these sensors can hone in on, Prasad said the applications are extensive.
“This imaging modality has been deployed aboard NASA aerial and satellite platforms, as well as the international space station,” he said. “It is continuing to play a critical role in earth science, for applications such as ecosystem monitoring, invasive species detection and the study of forest fires, to name a few.”