The Fabry-Perot interferometers (FPI) are essential components of many hyperspectral imagers (HSI). While the Piezo-FPI (PFPI) are still very relevant in low volume, high performance applications, the tunable MOEMS FPI (MFPI) technology enables volume-scalable manufacturing, thus having potential to be a major game changer with the advantages of low costs and miniaturization. However, before a FPI can be utilized, it must be integrated with matching optical assembly, driving electronics and imaging sensor. Most importantly, the whole HSI system must be calibrated to account for wide variety of unwanted physical and environmental effects, that significantly influence quality of hyperspectral data. Another challenge of hyperspectral imaging is the applicability of produced raw data. Typically it is relatively low and an application specific software is necessary to turn data into meaningful information. A versatile analysis tools can help to breach the gap between raw hyperspectral data and the user application. This paper presents a novel HSI hardware platform that is compatible with both MFPI and PFPI technologies. With an MFPI installed, the new imager can have operating range of λ = 600 - 1000 nm with FWHM of 15 - 25 nm and tuning speed of < 2 ms. Similar to previous imager in Ref. 1, the new integrated HSI system is well suited for mobile and cloud based applications due to its small dimensions and connectivity options. In addition to new hardware platform, a new hyperspectral imaging analysis software was developed. The new software used in conjunction with the HSI provides a platform for spectral data acquisition and a versatile analysis tool for a processing raw data into more meaningful information.
Skin cancers are a world wide deathly health problem, where significant life and cost savings could be achieved if detection of cancer can be done in early phase. Hypespectral imaging is prominent tool for non-invasive screening. In this study we compare how use of both spectral and spatial domain increase classification performance of convolutional neural networks. We compare five different neural network architectures for real patient data. Our models gain same or slightly better positive predictive value as clinicians. Towards more general and reliable model more data is needed and collection of training data should be systematic.