摘要
针对目前热轧带钢表面缺陷识别存在训练样本量小、识别效率低等问题,提出了一种基于脉冲卷积神经网络的带钢表面缺陷识别分类方法。为提高模型泛化性,首先利用扩散模型(Diffusion Model)对不平衡小样本数据集进行数据增强扩充,然后搭建脉冲卷积神经网络,并通过引入代理梯度方法进行网络监督训练,同时加入注意力模块来提高特征提取效率。实验结果表明:本文提出的脉冲卷积神经网络模型在保证识别率的基础上具有较强的生物合理性,为深度脉冲卷积神经网络在实际工程的应用提供借鉴。
A pulse convolutional neural network-based classification method for identifying surface defects on hot-rolled strip steel is proposed to address the issues of small training sample size and low recog-nition efficiency. To enhance the model’s generalization ability, a diffusion model is first utilized to augment and expand the imbalanced small sample dataset. Then, a pulse convolutional neural network is constructed and supervised training is conducted using a proxy gradient method. Addi-tionally, an attention module is introduced to improve feature extraction efficiency. Experimental results demonstrate that the proposed pulse convolutional neural network model not only ensures high recognition accuracy but also possesses strong biological plausibility, which provides valuable insights for the practical application of deep pulse convolutional neural networks in engineering.
出处
《建模与仿真》
2023年第6期5207-5217,共11页
Modeling and Simulation