Due to off-normal images,the number of alignment iterations will be increased and the collimation efficiency will be decreased.To overcome this difficulty,by analyzing the characteristics of off-normal images,a Bayesian classifier is designed to achieve image classification,and the distortion images are detected by the shape factor of beam image.Experimental results show that: by setting reasonable class condition and shape determination standard,the off-normal image classification and examination filtration can be effectively realized.As a consequence,the collimation cycle-index and alignment defeat′s probability due to the off-normal images is decreased,i.e.the total alignment efficiency is dramatically increased.In addition,the preliminary reason causing the off-normal images is analyzed,which supports the breakdown fast localization.