红外技术, 2018, 40 (9): 868, 网络出版: 2018-10-06
基于非线性模型的神经网络非均匀性校正方法
An Improved Neural Network Non-uniformity Correction Algorithm Based on Non-linear Model
低照度 短波红外 非线性模型 BP神经网络 非均匀性校正 low irradiance short-wave infrared non-linear model back-propagation neural network NUC
摘要
在低照度成像的短波红外相机中, 像元响应存在非线性问题。为了克服传统的神经网络自适应校正方法只能进行线性校正的不足, 提出了一种基于非线性模型的 BP神经网络非均匀性校正算法, 针对单一像元通过隐含层多神经元拟合像元校正曲线, 有效降低拟合误差, 并通过实验验证了算法的合理性。结果表明, 改进算法在图像的局部非均匀性, 粗糙度方面相较于传统算法分别降低了 27%和 28%, 非线性响应像元校正曲线拟合误差降为传统算法的 30%。
Abstract
In most short-wave infrared imaging systems working at low irradiance conditions, pixels work in a nonlinear manner. Considering that the traditional neural network correction methods are only applicable to linear responses, a new back propagation neural network non-uniformity correction (NUC) algorithm was proposed based on a nonlinear model. The three layers of neural networks were used to fit the correction curve of one single pixel, which effectively reduced the fitting error. Experiments were performed to verify the rationality of the proposed algorithm. The results demonstrated that it outperformed the traditional method on the local non-uniformity, roughness, and fitting error of the correction curve by 27%, 28% and 30%, respectively, for a single pixel.
程起森, 张元涛, 孙德新. 基于非线性模型的神经网络非均匀性校正方法[J]. 红外技术, 2018, 40(9): 868. CHENG Qisen, ZHANG Yuantao, SUN Dexin. An Improved Neural Network Non-uniformity Correction Algorithm Based on Non-linear Model[J]. Infrared Technology, 2018, 40(9): 868.