基于改进MobileNetV2的金属表面缺陷分类方法

Metal Surface Defect Classification Method Based on Improved MobileNetV2

摘要:

金属表面缺陷检测是工业制造中质量控制的关键环节；传统的人工检测方法由于成本高、效率低，难以满足现代制造业对高精度与高效率的需求；提出了一种基于MobileNetV2的改进网络模型，用于提高金属表面缺陷检测的精度与效率；在MobileNetV2网络基础上，引入坐标注意力机制以增强特征学习能力，采用深度可分离思想改进Inception模块，在增强网络对多尺度特征的提取能力的同时保持模型参数量；通过图像增强技术处理数据集，以提升网络的鲁棒性；实验在NEU-DET金属缺陷数据集上进行，验证了模型的有效性；IC_MobileNetV2模型在验证集上取得了92.8%的准确率，与原始的MobileNetV2、AlexNet、GoogleNet、DenseNet、ResNet34和ResNet50相比，准确率分别提高了5.6、2.8、0.9、1.7和1.7个百分点；实验结果表明，该方法在金属表面缺陷分类任务中具有较好的应用潜力。

Abstract:

Metal surface defect detection is a crucial aspect of quality control in industrial manufacturing; traditional manual inspection methods are costly and inefficient, making it difficult to meet the requirements of modern manufacturing for high precision and efficiency; a modified network model based on MobileNetV2 is proposed to improve the accuracy and efficiency of metal surface defect detection; the model introduces the Coordinate Attention (CA) mechanism to enhance feature learning ability and incorporates an improved lightweight Inception_DSC module to strengthen the extraction of multi-scale features; image augmentation techniques are applied to the dataset to improve the robustness of the network; experiments conducted on the NEU-DET metal defect dataset validate the effectiveness of the model; IC_MobileNetV2 achieves a validation accuracy of 92.8%, which is 5.6, 2.8, 0.9, 1.7, and 1.7 percentage points higher than MobileNetV2, AlexNet, GoogleNet, DenseNet, ResNet34, and ResNet50, respectively; experimental results show that this method has good practical significance in metal surface defect classification.