Abstract: The X-ray digital radiography automatic inspection system with single camera has been developed and successfully applied to the online automated test of some products containing complicated assembly structures by comprehensively considering multiple visual inspection technique and CT in order to solve the issue of nondestructive test of interior structures. First, a method, which uses the 3-point location of hardware and the integral projection of local area to find out a backward differential extremum, locates the product quickly. Second, the principal component entropy of the whole-orientation hierarchical hybrid projection of the ordered image set was extracted as the recognition feature basis. Finally, a BP neural network system was used to speed up the recognition process with parallel operations. The test process was automatically controlled by computer except loading and unloading. The experimental results show that wrong inspection ratio is less than 4.5% without missing inspection, and the average recognition takes less than 5s in test progress.

Key words: information processing, complex product, assembly structure, automatic recognition, system design