基于图像梯度的神经网络红外焦平面非均匀校正算法

红外焦平面阵列固有的非均匀性导致叠加在图像上的固定图形噪声严重影响了红外系统的成像质量。传统的神经网络非均匀校正算法存在待处理像素的期望值求解固有缺陷、收敛速度慢和学习速度过大，容易造成算法不收敛。提出了基于图像梯度的神经网络非均匀校正算法，通过对处理像素的期望值求解、改进和调整学习速度、改善图像校正效果,提高了算法收敛速度。通过对真实的红外图像序列实验表明,新算法相对传统的神经网络算法收敛速度提高了50%以上，红外图像校正效果也得到了提高。

Non-uniformity correction for IRFPA with image gradient-based neural network

The non-uniformity response in IRFPA detectors produces corrupted images with a fixed-pattern noise (FPN).There are several disadvantages in the traditional neural network based non-uniformity corrections, such as inherent defect in desired value of pending pixel, slow convergence speed and divergence because of high learning rate. In this paper, an image gradient-based neural network algorithm is presented, in which optimization techniques are used ,such as improvement in desired value of pending pixel and adaptive learning rate. The simulation with real infrared data sequence has proved that the new algorithm has the advantages of higher convergence speed and better correction effect.