Inspecting surface mounted devices using k nearest neighbor and Multilayer Perceptron

Automatic inspection of electronic components during the production of printed circuit boards is a way to ensure the quality of this production, reducing the cost of re-work. An automatic optical inspection system based on AI techniques for surface mounted devices is proposed in this work. The method relies on extracting the contour and histogram related features of component images, using Watershed segmentation, Canny edge detection, border following algorithm and histogram analysis. Histogram related features are applied in the k nearest neighbor technique with the goal of identifying the existence of a component. Contour related features are used to identify changes in angle and position by a comparison method and also to classify the component using a Multilayer Perceptron (MLP) neural network. Both techniques were used in the inspection system with the chosen features, and are validated through the 10-fold cross validation data method.