Artificial Neural Network based Temperature-robust Non-uniformity Correction for Infrared FPAs

Infrared focal plane array (FPA) detectors are prone to significant levels of spatial non-uniformity. As these detectors are sensitive to thermal signatures, the spatial non-uniformity fluctuates with ambient temperature conditions. If unaddressed, the spatial non-uniformity results in unpleasing noise in the infrared imagery. Majority of the prior works to address spatial non-uniformity have used handcrafted models to approximate FPA’s response and variations in non-uniformity. This work presents a learning based approach, using a simple artificial neural network, to address temperature varying spatial non-uniformity of infrared FPAs. The proposed method uses a novel FPA non-uniformity measure to track changes in non-uniformity due to ambient conditions. As shown by the experimental results, the proposed neural network clubbed with FPA non-uniformity measure, learned a suitable model for FPA response and non-uniformity variation. Also, unlike prior neural network based approaches which used CNNs, the proposed network is simple enough to be implemented on low-power hardware.