Optical Detection of Lightning via Diffractography

Recording spatial and spectral content has been the traditional domain of hyperspectral cameras. They produce very large datasets that require significant storage, transmission and processing capabilities. Despite the great usefulness of this information, they are captured in a very inefficient way; in many important problems, only a tiny subspace of the signal is necessary to produce the corresponding results. That is the case of classification or target detection problems. Using hyperspectral data with state-of-the-art machine learning algorithms (such as deep neural networks) has been a considerable challenge due to the sheer data sizes, immense hardware requirements and long training cycles.Our company has developed a technology that overcomes these hurdles. Our system captures spatio-spectral content in a compact, information-rich, monochrome image, termed diffractogram.nbsp; They can be used directly for inferencing, i.e., target detection and classification. This is achieved via optimally designed nanofabricated, diffractive-filter arrays (DFA) integrated into an existing sensors (FPAs).Compared to traditional approaches, we have identified strengths that can be exploited for a class of applications:Applicable for any spectral range (UV, Vis, NIR, SWIR, etc).Adaptable to existing sensors (virtually any FPA can be used)Small signals with a high density of information (efficient storage and transmission)Produced by optical compression (signals be used directly, no need for any decompression)Snapshot capture (no line scanning; well suited for moving targets)No moving parts (physically robust)Efficient at detecting temporal changesHigh transparency (no filters imply more light can be used)Spectral continuity (recorded signal is not spectrally discretized)Small size and weight (single optical path and no complex scanning components)Multi-purpose (single system produces data that can be used in many applications)