现代信息科技2025，Vol.9Issue(13)：7-12,6.DOI:10.19850/j.cnki.2096-4706.2025.13.002
基于小样本的微型扬声器振膜异常检测算法研究
Research on an Anomaly Detection Algorithm Based on Few-Shot of Micro-Speaker Diaphragms
摘要
Abstract
To address the inefficiency caused by manual visual inspection of micro-speaker diaphragms,this paper proposes an anomaly detection algorithm that integrates a Latent Diffusion Model and SimAM.The diffusion model is employed to synthesize samples,solving the common issue of insufficient samples in industrial scenarios.Additionally,SimAM is incorporated to enhance the algorithm's image feature extraction capability and overall performance.Experiments demonstrate that the proposed method achieves an accuracy of 96.57%on the validation set,outperforming other anomaly detection algorithms.Furthermore,the detection speed reaches 157 frames per second,meeting the industrial on-site inspection standards and real-time requirements.This provides support for anomaly detection in micro-speaker diaphragms.关键词
异常检测/小样本学习/注意力机制/特征融合Key words
anomaly detection/Few-Shot Learning/Attention Mechanism/feature fusion分类
信息技术与安全科学引用本文复制引用
李旭东,宋文龙,黄建平..基于小样本的微型扬声器振膜异常检测算法研究[J].现代信息科技,2025,9(13):7-12,6.基金项目
国家自然科学基金项目(61701105) （61701105）
