QUAD FLAT NO-LEAD (QFN) DEVICE FAULTY DETECTION USING GABOR WAVELETS

Computer vision inspection system using image processing algorithms have
been utilized by many manufacturing companies as a method of quality control. Since
manufacturing industries comprise of many types of products, various image processing
algorithms have been developed to suit different type of outputting products. In this
paper, we explored Gabor wavelet feature extraction as a method for vision inspection.
Unlike conventional vision inspection system which require manual human
configuration of inspection algorithms, our experiment uses Gabor wavelets to
fractionate the image into distinctive scales and orientations. Through chi-square
distance computation, the physical quality of Quad Flan No-Lead (QFN) device can be
distinguished by computing the dissimilarity of the test image with the trained database,
thus eliminating the weakness of human errors in configuration of vision systems. We
performed our algorithm testing using 64 real-world production images obtained from a
0.3 megapixel monochromatic industrial smart vision camera. The images consists a
mixture of physically good and defected QFN units. The proposed algorithm achieved
98.46% accuracy rate with the average processing time of 0.457 seconds per image.