Research On High Dynamic Welding Region Image Clarification And Information Fusion Based On Visual Characteristics

Welding is one of the most important material forming and metal working technologies in modern manufacturing.Intelligent welding is the future development direction,which will be influenced by clear acquisition of welding region images.Due to the great dynamic range of welding process,uneven illumination,and strong interference from arc and spatter,the existing vision sensors have the problem of underexposure or overexposure,so that the complete information of welding region can not be effectively obtained.Therefore,by considering the single image clarity and multi-exposure fusion,three methods are proposed to improve the dynamic range,obtain clear welding region images and realize the information fusion of welding region with high dynamic range.The specific research contents are as follows:(1)Firstly,a JND model-based method for the clarity of high dynamic welding region image is proposed.A JND saliency detection model based on brightness masking and texture masking is established to obtain the saliency map of welding region,which can effectively detect the key information of welding region.Then,a contrast enhancement algorithm is proposed to solve the problem of low contrast after the saliency reversal,which is different from the traditional enhancement algorithm by stretching the histogram.The proposed contrast enhancement algorithm not only can achieve the purpose of image enhancement,but also avoid information loss.The experimental results show that the proposed method can effectively overcome the strong interference,expand the dynamic range,and obtain a clear welding region image.(2)Secondly,a multi-exposure welding region image information fusion method based on Laplacian pyramid is proposed.Considering the advantages of multiscale transform,a multi-scale hybrid weight is proposed by employing the saliency weight from spectral residual model,the local exposure weight of single image and global exposure weight between different exposure images,which can preserve the useful information of all multi-exposure welding images and abandon redundant information simultaneously.Then,a fusion guideline for multi-exposure welding image fusion is proposed,by which a clear and complete welding region information can be fused under the guidance of hybrid weight.In addition,the pyramid reconstruction process is boosted by applying the guiding image filtering that is also performed before the saliency map,which effectively reduces artifact noise and preserves the edge detail information.(3)Thirdly,a multi-exposure welding region image information fusion method based on Haar wavelet gradient reconstruction is proposed.The X and Y gradient maps for the multi-exposure welding image are calculated by applying the Hudgin gradient model.Then,all gradient maps of X and Y direction be fused to obtain the fusion gradient map according to the maximum gradient magnitude.By applying the Haar wavelet gradient reconstruction algorithm with Poisson solver,the artificial noise in the reconstruction process can be effectively avoided because the fusion gradient does not satisfy the conservative field conditions.For the multi-exposure color image fusion,the source images are firstly converted to the YCbCr color space.The luminance channel Y is fused by applying the above proposed method,and the chrominance channel Cb and Cr is fused by the weighted average principle.It will be converted to RGB color space to obtain the final fused image.(4)Finally,the validity of the proposed method is verified by the subjective and objective evaluation.Furthermore,the proposed two multi-exposure fusion method not only can adapt to welding image with great dynamic range,but also can be applied to other industrial fields.Experimental results also show that the fusion performance of proposed method outperforms the previous methods in the multi-exposure color image fusion field.