Circuit board electronic element target detection method based on simplified YOLOv3 network.

Abstract

The invention provides a circuit board electronic element detection method based on a simplified YOLOv3 network. A small number of original images are intercepted according to different starting points in a preset size to acquire a large number of training samples. A dual-model parallel detection mode is adopted, one model detects blue annular resistors, the other model detects capacitors, then detection results of the two models are summarized, a non-maximum suppression method is adopted to eliminate a redundant detection window, and the mode solves the problem that the data volume of target objects with more blue annular resistors and less capacitors in original data is unbalanced. And the detection accuracy is improved. Images for training and identifying the blue annular resistance model are subjected to color filtering, the blue part is reserved, the target features are highlighted, and identification accuracy is improved. The YOLOv3 is cut, the weight number is reduced, the calculation cost is reduced, and the detection efficiency is improved. The category, the position and the size information of the detection target are automatically output through the model, and the problem that related AOI equipment can only carry out defect detection through manual processing and element marking is solved.

Description

【Technical field】 [0001] The present invention relates to the technical fields of artificial intelligence, optics, computer vision, and automation control technology, and in particular to a detection method, device, terminal, and storage medium for circuit board electronic components based on a simplified YOLOv3 neural network. 【Background technique】 [0002] In industrial production, it is usually necessary to use AOI (Automated Optical Inspection, automatic optical inspection equipment) to detect electronic components after PCB (Print Circle Board, printed circuit board), such as blue ring resistors, ring capacitors, block Shape capacitors, etc., and then judge whether there are defects such as wrong insertion, reverse polarity insertion, damage, and component bending. Specifically, it is usually necessary to do the manufacturing process first, manually mark the electronic components in the pictures taken by AOI, and then intercept the marked electronic components from the...

AI Technical Summary

Problems solved by technology

[0005] The purpose of the present invention is to provide a detection method, device, terminal and storage medium for circuit board electronic components based on a simplified YOLOv3 neural network, aiming to solve the problem of insufficient recognition accuracy of the trained neural network model due to fewer PCB images. high problem

Images