Design of A Detection and Recognition System for Mounting Defects of Surface Mount Technology Components Based on Improved Convolutional Neural Networks

The manuals or traditional image recognition methods are widely adopted on the detection for mounting of surface mount technology (SMT) components at present though they are low in efficiency and detection accuracy. Thus, a detection and recognition system for mounting defects of SMT components based on the improved convolutional neural networks (CNN) was designed in this paper. The image registration was realized on the test image by the algorithm of accelerated robust feature in the system. The test images after binarization were contrasted by differential calculation to find out the area of defects. An improved model of CNN which was based on traditional models of CNN was designed to identify mounting defects. A device for image acquisition with a dual-light source and adjustable brightness was designed, and the interface for human-computer interactions was developed based on C++. The actual measurements proved that the detection and recognition system designed in this paper well overcame the influence of ambient lights on the stability of image acquisition, and the accuracy of recognition was greatly improved compared with existing methods for mounting defects detection and recognition. The system has certain practicability and market value.