A Study Of The Non-uniformity Correction For Infrared Focal Plane Arrays

Confined by the materials and technological level, the IRFPA (Infrared Focal Plane Arrays) has the character of non-uniformity in each sensing unit, which results in the Fixed Pattern Noise and has a bad effect on the infrared image. Therefore, it is necessary to correct the non-uniformity for the focal-plane thermal imaging systems.This paper firstly introduces the definition of non-uniformity, analyzes the two kinds of response model of the detectors and presents the common non-uniformity correction method. Because of narrow dynamic range, the two-point correction method based on the linear response model cannot satisfy the need in practice. According to the nonlinearity of the detector response model, the polynomial least-square fitting method is applied to the non-uniformity correction, and a novel multi-point correction based on the'S'curve model is put forward. The new method enlarges the dynamic range and improves the precision of the correction.At the same time, this paper also analyzes the classic BP neural network correction method. In consideration of both hide layer and the selection of the training initial values, two new correction methods are introduced. Because that the simple mean filter will blur the image, the concept of the membership of region has been introduced into the hide layer algorithm, and using the gradient calculate the desired output value. This method can keep the detail of the image edge. Another method using the two-point method as the training initialization has quicker convergence speed and higher precision.