Infrared Image Non-uniform Correction Algorithm Based on Improved CNN Network Model
Aiming at the problems of the existing non-uniform infrared image correction algo-rithms,a correction algorithm based on the improved CNN network model is designed.Firstly,the col-lected infrared images are registered and denoised to form a two-dimensional infrared image dataset as the input item of model.The CNN model is constructed and suitable convolution kernel and step size is selected for the convolution layer.In order to suppress the gradient dispersion of the convolution layer and further improve the training ability of two-dimensional infrared image data,residual blocks are used to optimize and improve the convolution layer.Finally,the edge of the fused infrared image is cor-rected based on the least mean square algorithm.The experimental results show that the proposed non-uniform correction algorithm can effectively improve the problems of uneven brightness and noise in the image,and the mean roughness and mean root mean value of the 5 regions of the corrected image are 1.779 and 0.643,respectively,which are significantly improved compared with the original image,and the correction effect is better than the two traditional algorithms.
万方数据
维普
