Non-uniform system response detection for hyperspectral imaging systems

Abstract

Near infrared (NIR) hyperspectral imaging (HSI) has established itself as a powerful non-destructive tool for the chemical analysis of heterogeneous samples. However, one of the main disadvantages of NIR HSI is that the technique suffers from instrumentation-related problems, which in turn affect the acquired images. In general, focal plane array (FPA) based hyperspectral systems are affected by spatial and spectral non-uniform response, the presence of defective sensors (e.g. dead or saturated sensors), and temporal and spatial (e.g. dark current) noise. Another issue is each new camera system needs to be calibrated to assess its specific responses to light. To correct for these issues, we used known standards to measure the response of the sensors and capture the location of the field of view and defective sensors using linear and quadratic models. The parameters of these models were then used as input features for classification of sensor responses using a k-means algorithm. The results conclude that linear models are insufficiently precise for calibration but estimate sufficiently accurately the system’s response and functionality. Specifically, it was shown that the classification method discriminates non-responsive regions effectively.