Although laser welding has high efficiency , it is difficult to operate. The welded workpiece is prone to defects , incomplete penetration or excessive penetration , leading to unqualified welding quality. Facing this situation , in order to realize the effective inspection of the welding workpiece , it is very necessary to implement the laser weld penetration monitoring. In the research , the height camera is used to collect the plasma topography image and carry out four pre-processing to remove the interference information in the image. Four kinds of feature information of plasma to- pography image are extracted , including centroid width , centroid height , centroid swing angle and plasma area. Based on BP neural network , the mapping relationship model between the ionomer feature information and the laser weld pen- etration type is constructed. With the feature information as the input , the penetration probability value is output. The type whose probability value is closer to 1 is the monitored penetration result.