Research On Multifaceted Microscopic Lmaging And Intelligent Defect Recognition Technology For Microcapacitor

In response to the outdated technology in micro-capacitor defect recognition and the long-term reliance on imported equipment in China,this paper presents research on an artificial intelligence-based algorithm for identifying surface defects in micro-capacitors.A prototype of a micro-capacitor defect recognition device was developed,achieving good compatibility with the micro-capacitor production process and filling a significant gap in domestic technology.Firstly,based on the image characteristics of micro-capacitor surface defects,this study explores the application of YOLOv8-based high-accuracy improvement and zero-shot learning methods in micro-capacitor defect recognition.Secondly,according to the algorithm requirements,an optical enhancement machine vision system for micro-capacitor surface imaging was studied,leading to the successful development of a system for micro-capacitor surface defect recognition.Finally,experiments were conducted using the micro-capacitor surface defect recognition system,establishing a micro-capacitor defect dataset,and verifying the effectiveness of the YOLOv8 high-accuracy improvement and zero-shot learning methods,thereby improving defect recognition accuracy.Through this research,the technical bottleneck of micro-capacitor surface defect recognition was overcome,resulting in the successful development of the first domestic prototype of a micro-capacitor defect recognition device.The main contributions of this paper include the following four parts:(1)A high-accuracy method based on YOLOv8 for micro-capacitor defect recognition is proposed.By integrating a similarity attention mechanism,this method effectively focuses on tiny defects in the image.Additionally,by optimizing the multi-scale information fusion strategy and introducing the Wise-IOU loss function,the accuracy and robustness of defect recognition are improved.Experimental results show that the average precision of this method reaches 95.8%,an improvement of 9.5%compared to before optimization.(2)A zero-shot learning defect recognition method based on dual variational autoencoders and generative adversarial networks is proposed.The combination of dual-branch autoencoders and generative adversarial networks,along with the introduction of cross-alignment loss and distribution alignment loss functions,enhances the performance and generalization ability of zero-shot learning.Experimental results show that the accuracy of this method in generalized zero-shot learning for seen classes,unseen classes,and their harmonic mean values reached 61.3%,65.5%,and 63.3%respectively.(3)The development of micro-capacitor defect recognition equipment based on intelligent algorithms for microelectronic components was completed.First,the overall conceptual design of the system was carried out,optimizing the motion imaging system and product supply system to form a logically clear product detection process.Secondly,the mechanical structure of the system was designed,developing a five-sided imaging technology that improved imaging clarity.Thirdly,control software based on PLC was developed,realizing full-process functionality from sample feeding to imaging sorting.Finally,system assembly,debugging,and intelligent algorithm program deployment were completed.(4)The testing and verification of micro-capacitor defect recognition were completed.Using the micro-capacitor defect recognition system,samples from multiple manufacturers were tested,resulting in the construction of a micro-capacitor dataset containing 1,377 images.The improved YOLOv8 high-accuracy method and proposed zero-shot learning method were used to complete defect recognition tests,meeting the actual needs of the micro-capacitor production line.In summary,based on innovative high-accuracy methods of YOLOv8 and zero-shot learning methods utilizing dual variational autoencoders and generative adversarial networks,this paper successfully developed a prototype system for micro-capacitor defect recognition,filling a critical domestic gap in this field.