基于改进CBAM注意力机制的MobileNetV3风扇异常状况识别研究  

MobileNetV3 Fan Anomaly Recognition Based on Improved CBAM Attention Mechanism

作　　者：刘明 王荣燕[1] 王汝旭 武高旭 张佳宁 梁俊祥 

机构地区：[1]德州学院,山东德州253023

出　　处：《工业控制计算机》2025年第3期90-92,共3页Industrial Control Computer

基　　金：省级大学生创新训练项目(202310448083)。

摘　　要：工业风扇在生产设施中起着至关重要的作用,关键风扇的突然停机对安全生产影响巨大。通过分析在-6 dB噪声环境中的故障风扇发出的声音,提取声音样本的语谱图,采用MobileNetV3模型,针对该模型注意力模块SE(Squeeze-and-Excitation)存在的参数化程度较低问题,采用空洞卷积(Dilated Convolution)优化的卷积块注意力模块CBAM(Convolutional Block Attention Module)予以替代,提出了改进后的MobileNetV3模型。实验结果显示,该模型的分类准确率达到了98%,相较于原MobileNetV3模型,准确率提升了2.07个百分点。The industrial fan plays a crucial role in production facilities,and the sudden shutdown of key fans can have a significant impact on safety and production.This paper analyzes the sounds emitted by faulty fans in a-6 dB noise environment,extracts spectrogram of the sound samples,and uses the MobileNetV3 model.To address the low parameterization of the SE(Squeeze-and-Excitation)attention module in this model,the study replaces it with the convolutional block attention module(CBAM)optimized with dilated convolutions.The improved MobileNetV3 model achieves a classification accuracy of 98%,representing a 2.07 percentage point increase compared to the original MobileNetV3 model.

关 键 词：空洞卷积 CBAM MobileNetV3 迁移学习 SPECTROGRAM 

分 类 号：G63[文化科学—教育学]