基于卷积神经网络和模态转换的磁瓦内部缺陷检测方法

Internal defect detection of magnetic tile based on CNN and modal transformation

针对人工检测磁瓦内部缺陷过程中需要成熟的经验知识,检测过程不稳定且效率较低等问题,设计一套智能化检测系统。受人工检测的启发,提出一种基于卷积神经网络和模态转换的磁瓦内部缺陷检测方法。将时域信号转换为时-频域语谱图,利用卷积神经对语谱图提取特征并分类。为更精准地强调重要信息而抑制无关信息,将坐标注意力机制引入到卷积神经网络中。提出的基于卷积神经网络和模态转换的预测模型的准确率达到98.4%,证明提出的检测方法对于磁瓦内部缺陷检方法是有效的。实验结果表明,模态转换和坐标注意力机制能提升模型的性能。

Aiming at the demand for mature experience knowledge,unstable detection process and low efficiency in the processing of manual operation,an intelligent detection system is designed to avoid those drawbacks.Inspired by manual detection,we propose an internal defect detection method of magnetic tile based on convolution neural network(CNN)and modal transformation.The time domain signal is transformed into time-frequency domain spectrogram,and the convolution neural network is used to extract features and classify the spectrogram.In order to precisely emphasize important information and suppress irrelevant information,the coordinate attention mechanism is introduced into CNN.The accuracy of the prediction model based on convolution neural network and modal transformation achieves 98.4%,which proves that the proposed detection method is effective for the internal defect detection method of magnetic tile.The experimental results show that the modal transformation and coordinate attention mechanism can improve the performance of the model.