Fig. 1. Flow chart of algorithm

Fig. 2. Flow chart of training and classification process of convolutionneural network

Fig. 3. Output results of single part. (a) Test result of "battery"; (b) test result of "gear"

Fig. 4. Angle matching process between testing workpiece projection and standard workpiece projection

Fig. 5. Internal part definition of homemade parts

Fig. 6. Training set of projection data

Fig. 7. Test result of qualified workpiece

Fig. 8. Test result of unqualified workpiece which missing internal parts

Fig. 9. Detection results under two angles. (a) Misdetection; (b) detection under 90° rotation

Fig. 10. Detection results of internal parts missing of fuze workpiece. (a) Upper test result; (b) middle test result; (c) lower test result

Fig. 11. Current-angle detection image matching standard workpiece image at same angle. (a) Classification result; (b) matching result; (c) displacement detection

Fig. 12. Contrast of feature selection results of proposed method and previous methods. (a) Regional features of previous methods; (b) individual characteristics of proposed method

Fig. 13. Detection result after rotation of 5°