Multi-source Molten Pool Image Information Fusion

Welding is indispensable in industrial production process.With the proposal of industrial 4.0 and the rapid development of artificial intelligence,using computer vision to monitor the welding process will be one directions of future development.However,due to the strong arc light interference in the welding process,the image quality of weld pool can not meet the requirements of computing.The image of weld pool obtained by traditional single sensor can not reflect the whole picture of weld pool.Based on this application requirement,this paper research on image fusion of multi-source welding pool.Aiming at the two different process requirements of Cold Metal Transferring Mixed With Pulses(CMT+P)and Tungsten Inert Gas Welding(TIG),two sets of different pool imaging systems are studied and designed.Two different fusion algorithms of multi-source weld pool images are proposed.(1)CMT+P fusion algorithm based on multi-band weld pool image.In order to solve the problem that the current weld pool imaging device has only a single band.,we designed a multiband imaging device.The gray-scale images in 850 nm high pass and 660 nm narrow band and full-color image of weld pool were acquired.We proposed a detail enhanced multi-band fusion algorithm for weld pool image.The dual band gray pool image is decomposed by NSCT.After extracting the gradient features of the low-frequency coefficients through the nonnegative matrix decomposition in gradient domain,we obtain the adaptive fusion coefficient matrix of low frequency fusion.The high frequency fusion coefficient based on the spatial frequency of neighborhood.On the basis of gray fusion results,an improved Reinhard algorithm is proposed to fuse the color information..The experimental results show that compared with other fusion algorithms,our fusion algorithm can produce a natural color,leading to an outstanding fusion results.(2)HDR generation enhancement fusion algorithm based on multi-exposure TIG weld pool image.In order to solve the problem that the current image acquisition of weld pool is difficult to take both arc and weld pool area into account,we designed a weld pool device of image acquisition with high and low exposure.We also proposed a fusion HDR generation algorithm based on image enhancement of weld pool with high and low exposure.The detail layer of image enhancement is constructed by improved domain-guided filtering,and the basic layer of image is constructed by contrast,exposure and saturation factors.The halo is eliminated by recursive filtering,the basic layer and detail layer are enhanced fused to achieve the high and low exposure image fusion.The experimental results show that the proposed HDR generation algorithm effectively fuses the complementary information between high and low exposure pool images,eliminates arc redundancy,improves the detail contrast of pool images,and conforms to the observation habits of human eyes.