一种改进的神经网络红外焦平面阵列非均匀性校正算法

An Improved Nonuniformity Correction Algorithm for IRFPA Based on Neural Network

传统的神经网络非均匀性校正算法对噪声具有较好的自适应性,但当空间低频噪声较大时,校正效果明显下降。为此,提出了一种传统神经网络同场景的一阶统计相结合的改进算法。将对偏置的估计转化成对场景的统计和对辐射均值的估计,新算法较原算法具有了更强的校正能力,特别适合于非均匀性主要由偏置产生的焦平面器件。理论分析和比较实验结果显示了其优越性。

The traditional neural network correction has a good adaptivity to the noise. But with a stronger low frequency space noise, the correction effect is very poor. So an improved algorithm integrated the traditional neural network with the first-order statistics of scene has been proposed, which can convert the offset estimation into the estimation of radiation average and scene statistics. As compared with old algorithm, new algorithm has stronger correction ability, and is specially adapted for the infrared focal-plane array （IRFPA） that the offsets dominate nonuniformity. Theoretical analysis and the experiment with real infrared data show that the proposed algorithm has advantages over others.