Research On Chip Component Recognition Technology Based On Deep Learning

The complexity,precision and integration of modern circuit board design have put forward higher requirements for circuit board detection technology,and the development of SMD components has also experienced changes from large to small,from simple to precision.Identifying the SMD components can improve the detection accuracy of the circuit board,and the recognition of the printed characters on the surface of the SMD components further consolidates the detection robustness.Based on the existing circuit board automatic detection system,this paper analyzes and researches the core algorithm in the detection process,that is,the patch component identification technology.Through the research on the deep learning algorithm,the Yolo v3 and Res Net networks are improved,and the patch element The accuracy of device classification and character recognition has been verified by the self-made data set,which speeds up the detection speed,meets the real-time requirements,and improves the ability of the circuit board to identify SMD components.In this paper,the deep learning algorithm is used to identify the SMD components in the ordinary background and the circuit board background,and the recognition research is carried out on the two parts of target detection and character recognition.Main research content: This paper improves the patch component recognition algorithm based on Yolo v3.The algorithm first uses the K-means algorithm to re-cluster the self-made data set to generate Mounted anchors,and then introduces the lightweight network Mobile Net v3 as Yolo.The backbone network of v3 extracts the features of the input pictures,strengthens the ability of target detection through multi-scale feature fusion,and compares the ability to identify patch components with the classic algorithms.After experimental comparison and system testing,the accuracy rate reaches 89.36%.In order to reduce the false detection rate,in this paper,a new network model SS＿Res Net-50 is obtained by introducing the Spatial transformer network（STN）and channel attention mechanism to the character recognition algorithm based on Res Net,and using STN to correct the deformed text characters,Then,the character features are extracted by the improved network model to realize character recognition.After testing with multiple data sets,the recognition accuracy of the improved algorithm is improved.The research in this paper improves the detection efficiency and accuracy of the circuit board automatic detection system,and provides ideas and research directions for promoting the development and exploration of circuit board detection technology.