基于多尺度CNN的光伏组件缺陷识别

PHOTOVOLTAIC MODULE DEFECT IDENTIFICATION BASED ON MULIT-SCALE CONVOLUTION NEURAL NETWORK

为提高光伏组件缺陷分类精度与效率,提出一种改进的多尺度卷积神经网络模型(IMCNN)。该算法根据光伏组件缺陷特点,构建3个不同尺度端对端的卷积神经网络模型,同时为优化网络结构,在3个通道中均引入SE-Inception模块。首先由多通道卷积提取精密度不同的特征;再将这些特征进行融合,得到特征的增强表达;最后实现光伏组件的缺陷分类。由于光伏组件的缺陷样本较少,使用生成对抗网络生成一部分图像样本,达到有效进行数据增强的目的。实验结果表明,所提算法的Kappa系数较高,分类精度与效率均有明显提升。

In order to improve the accuracy and efficiency of photovoltaic module defect classification,an Improved Multi-scale Convolutional Neural Network(IMCNN) model is proposed. According to the defect characteristics of photovoltaic module,the algorithm constructs the end-to-end Convolutional Neural Network model of three channels with different scales. At the same time,in order to optimize the network structure,SE-Inception modules are introduced in all three channels. Firstly,multi-channel convolution is used to extract features with different fineness. Then these features are fused to obtain the enhanced expression of high-level features.Finally,the defect classification of photovoltaic module is realized. Due to the small number of defective samples of photovoltaic module,a part of the image samples is generated using the Generative Adversarial Networks to effectively expanding the dataset. The results show that the Kappa coefficient of the proposed algorithm is higher,and the identification accuracy and efficiency are significantly improved.