Computer Vision Model for Semantic Segmentation of Electronic Components in 2D Space

We develop a vision system, based on ENet deep convolution network, for recognition, classification, localization and boundary contouring of three types of electronic components. We acquire input images using a 6-axis industrial robot manipulator and a Raspberry Pi, ensuring appropriate communication between them. After pre-processing, input data is augmented to increase their number 40 times. Annotation is carried by a semi-automatic tool to obtain dataset pairs. Transfer learning is conducted using pre-trained ENet state and results are provided and analyzed.