摘要
在应用神经网络进行光伏组件红外图像故障检测中,针对其存在学习效率低、运行速度慢等问题,提出一种改进InceptionV3网络的故障分类方法。首先,通过设置宽度因子来压缩模型通道数;然后通过SENet机制实现网络结构优化,提升故障特征的提取能力;最后通过引入logcosh函数为损失函数增加约束项,确保输出loss值的稳定性并解决数值溢出问题。实验结果表明,改进后的InceptionV3模型在提升准确率的同时减少了运行时间,为光伏组件的故障分类提供了一种有效方案。
In the application of neural network in infrared image fault detection of photovoltaic module,aiming at the problems of low learning efficiency and slow running speed,a fault classification method based on improved InceptionV3 network is proposed.First,the number of model channels is compressed by setting the width factor;Then the network structure is optimized through the SENet mechanism to improve the ability of fault feature extraction;Finally,the logcosh function is introduced to increase the constraint term for the loss function to ensure the stability of the output loss value and solve the numerical overflow problem.The experimental results show that the improved InceptionV3 model improves the accuracy and reduces the running time,which provides an effective scheme for the fault classification of photovoltaic modules.
作者
姜萍
李梦瑶
栾艳军
JIANG Ping;LI Mengyao;LUAN Yanjun(College of Electronic Information Engineering,Hebei University,Baoding 071000,China)
出处
《激光杂志》
CAS
北大核心
2022年第8期90-94,共5页
Laser Journal
基金
河北省自然科学基金(No.A2020201021)。