Abstract: In the process of laser welding, the plume and spatter images contain plenty of information, which can directly reflect the welding status. During bead-on-plate disk laser welding of type 304 austenitic stainless steel plates, a high-speed camera in ultraviolet band and visible light band was applied to capture the plume and spatter images. Angular second moment, inertia moment, entropy and correlation of an image were calculated to describe the GLCM(gray level co-occurrence matrix) which is the second statistical characteristic values of regional texture, and analyze the inherent law between GLCM and the weld formation. The numbers and areas of spatter, orientation of centroid of plume,height of plume were calculated as the characteristic parameters. Weld formation was predicted by establishing a BP (back propagation) neural network model. Experimental results show that the proposed method can reflect the relationship between plume and spatter and welding status, and provide a basis for monitoring and control of high-power disk laser welding process in real time.