Computer Vision News - February 2023

30 Spectral Imaging Tool Welcome to new and old readers of Computer Vision News! This month, we review something quite different than usual. In fact, we are going to talk about a domain-specific deep learning tool, named SpectrAI, solely dedicated to the analysis of spectral data through neural networks. SpectrAI includes both command line and graphical user interface (GUI) options. It’s fully written in the well-known Pytorch library (Python), but thanks to the Matlab-based GUI, it is also well-suited to deep learning newbies who can select models and hyperparameters much easier than through code. But what kind of data does SpectrAI analyse? Not really the images we commonly discuss in this section of the magazine! This technique produces 1D(a spectrum, λ) or 3Ddata (a spectral hypercube, x×y×λ) obtainedbyacquisition at different wavelengths. A spectral dataset can be thought as either a collection of images at a specific wavelength stacked together or as a collection of wavelengths at different pixel locations. Moreover, spectral imaging is not only used in the biomedical imaging field, but it has also been applied to environmental monitoring, materials characterisation, and much more. However, given its atypical nature, spectral images might not be suitable for standard techniques we apply to the analysis of other medical images. That’s why SpectrAI provides numerousbuilt-inspectral datapre-processingandaugmentationmethods, neural networks for spectral data including spectral (image) denoising, spectral (image) classification, spectral image segmentation, and spectral image super-resolution. By Marica Muffoletto (twitter) A PEAK INTO THEWORLD OF SPECTRAL IMAGINGWITH SPECTRAI