Assembly of electronic components in printed circuit boards(PCBs) is prone to errors which require timely identification, which is usually done through automated optical inspection (AOI) methods. Recent developments have proposed the use of convolutional neural networks (CNNs) to perform inspections in industrial settings. CNN however have a limitation in which a large amount of data is required, which is particularly challenging for electronic component defect recognition, since defective samples are more difficult to obtain. In this work we propose a new CNN architecture for electronic component defect classification, using a two stage transfer learning model training approach to deal with the low amount of data problem. We tested our approach on a dataset of electronic components for defect classification, obtaining 99.45% accuracy, outperforming other popular neural network models.