An Inspection Method of SMD Component Type Based on Moment Features and BP Neural Network

Abstract:

The type inspection of surface mounted devices (SMD) components is an important part of the numerical control (NC) placement machine’s vision system. In order to improve the detection speed, accuracy rate and versatility, a detection method based on moment features and neural network is proposed. Firstly, component images are preprocessed in order to eliminate the influence of non-uniform illumination and simplify the calculation, so component lead images can be obtained, and then seven invariant moments and a zero-order Zernike moment of the lead images are extracted. Next, the moment features are corrected and normalized. Finally, back propagation (BP) neural network based on the Levenberg-Marquardt algorithm is taken as a classifier for training and testing the 8-dimensional mixed moment feature vectors, 0 and 1 are used to represent the degree of belonging of each image. The experimental results show that this method doesn’t need complex lighting system and has good versatility, and the correct rate can be up to 100%.

• Periodical: Applied Mechanics and Materials (Volume 610)
• Pages: 296-301
• DOI: https://doi.org/10.4028/www.scientific.net/AMM.610.296
• Online since: August 2014
• Authors: Zu Jin Wang, Xiao Diao Huang*, Ping Ping Gu
• Keywords: BP Neural Network (BPNN), Invariant Moment, Type Detection, Vision System, Zernike Moment

References

[1] Byung M K, and Hyung S C. Neural network-based parts classification for SMT processes, Proceeding of SPIE-The International Society for Optical Engineering, 2001, 4564: 268-279.

[2] Guan B Q, Yu X R, and Wang S G. An inspection method of component type based on moment features and chain code, Journal of Shanghai Jiaotong University, 2005, 39(6): 969-974.

[3] Fan L N, and Xu X H. Fuzzy classification based on moment invariant feature and neural networks for image pattern, Journal of Northeastern University (Natural Science), 2004, 25(8), 738-741.

[4] Mohamed R, Haniza Y, Puteh S, et al. Object Detection using Geometric Invariant Moment, American Journal of Applied Sciences, 2006, 2(6): 1876-1878.

[5] Zhang H, Shu H Z, Haigron P, et al. Construction of a complete set of orthogonal Fourier–Mellin moment invariants for pattern recognition applications, Image and Vision Computing, 2010, 28(1): 38-44.

DOI: 10.1016/j.imavis.2009.04.004

[6] Qin Z. Vehicle classification based on image moment invariant feature and BP neural network, Journal of South China University of Technology (Natural Science Edition), 2009, 37(2): 87-91.

[7] Papakostas G A, Karakasis E G, Koulouriotis D E. Novel moment invariants for improved classification performance in computer vision applications, Pattern Recognition, 2010, 43(1): 58-68.

DOI: 10.1016/j.patcog.2009.05.008

[8] Hu M K. Visual pattern recognition by moment invariants, IRE Transactions on Information Theory, 1962, 8(2): 179-187.

DOI: 10.1109/tit.1962.1057692

[9] Li S, Lee M C, Pun C M. Complex Zernike moments features for shape-based image retrieval, Transactions on Systems, 2009, 39(1): 227-237.

DOI: 10.1109/tsmca.2008.2007988

[10] Revaud J, Lavoue G, Baskurt A. Improving Zernike moments comparison for optimal similarity and rotation angle retrieval, Transactions on Pattern Analysis and Machine Intelligence, 2009, 31(4): 627-636.

DOI: 10.1109/tpami.2008.115

[11] Zhu H Q, Liu M, Ji H J. Combined invariants to blur and rotation using Zernike moment descriptors, Pattern Analysis and Applications, 2010, 13(3): 309-319.

DOI: 10.1007/s10044-009-0159-9

[12] Wang J M, Qin W Z. BP neural network classifier based on levenberg-Marquardt algorithm, Geomatics and Information Science of Wuhan University, 2005, 30(10): 928-931.