Yang F.-C.Kuo C.-H.Wing J.-J.Yang C.-K.CHUNG-HSIEN KUO2022-05-242022-05-242004https://www.scopus.com/inward/record.uri?eid=2-s2.0-15744399883&doi=10.1109%2fICSMC.2004.1400799&partnerID=40&md5=bd469b7edb26a3219b1aea9e641a7a2ahttps://scholars.lib.ntu.edu.tw/handle/123456789/611636In general, the laser Inspection can measure accurate 3D solder paste surface model, however, it is not practical due to the high cost and low inspection speed. This paper presents the three-dimensional (3D) solder paste surface model reconstruction using the image processing and artificial neural network (ANN), and the proposed approach forms the virtual laser 3D automatic optical inspection (AOI) model. The input nodes of the ANN model consist of the image features that are captured from images of using different light sources. The output nodes are the heights of the corresponding image pixel areas. The training patterns of the proposed ANN model use the laser 3D inspection results. Meanwhile, the in-lab design and the commercial coaxial light sources with the pad and sub-area based learning architecture models are constructed and validated, and the estimated 3D surface model achieves 90% accuracy in average. ? 2004 IEEE.Automatic optical inspection (AOI)Learning architecture modelsSolder paste inspectionThree dimensional (3D) reconstructionComputer architectureComputer visionFeature extractionImage processingImage reconstructionLaser beamsLearning systemsMathematical modelsThree dimensionalVirtual realityNeural networks[SDGs]SDG9conference paper10.1109/ICSMC.2004.14007992-s2.0-15744399883