基于轻量化深度学习的热浸镀Zn-Al-Mg镀层腐蚀强度检测  

Corrosion strength detection of hot dipped Zn-Al-Mg coating based on lightweight deep learning

作　　者：李丹 付国帅 Li Dan;Fu Guoshuai(College of Information Engineering,Zhengzhou University of Industrial Technology,Zhengzhou 451150,China;Hebi Institute of Engineering and Technology,Henan Polytechnic University,Hebi 458030,China)

机构地区：[1]郑州工业应用技术学院信息工程学院,河南郑州451150 [2]河南理工大学鹤壁工程技术学院,河南鹤壁458030

出　　处：《电镀与精饰》2025年第4期71-77,112,共8页Plating & Finishing

基　　金：教育部产学合作协同育人项目“基于海豚实验平台的人工智能专业‘双师型’教师培训实践探索”(230705384254049);河南理工大学鹤壁工程技术学院2024年度校级教育教学改革研究与实践项目“竞赛助力,强化实践,产教融合,培养计算机专业应用型创新人才(2024-JGZD-001)”;“新工科背景下计算机类专业校企协同育人模式研究(2024-JGYB-004)”。

摘　　要：热浸镀Zn-Al-Mg镀层腐蚀过程中产生的多种腐蚀物质,导致表面图像包含锈斑、划痕、油污等大量的噪声和干扰信息,以深度学习为主的人工智能检测方法会陷入复杂的循环迭代过程,效率过低。为了提升热浸镀Zn-Al-Mg镀层腐蚀强度检测的精确性、效率及降低计算资源消耗,提出基于轻量化深度学习的热浸镀Zn-Al-Mg镀层腐蚀强度检测方法。采用提升小波技术设计自适应提升机制,利用提升小波的多尺度分析和去噪能力,对腐蚀钢板图像进行预处理,提取出更清晰的腐蚀特征。将自适应提升机制嵌入到卷积神经网络的初始层级,构建卷积神经网络-ALS镀层腐蚀强度检测模型。为了进一步降低检测的复杂性,利用时延约束下的模型轻量化方法对构建模型进行轻量化处理,通过求解轻量化模型,实现热浸镀Zn-Al-Mg镀层腐蚀强度检测。测试结果表明:设计方法对于较小及微小的腐蚀情况的强度检测准确,显著减少了实施轻量化处理后的镀层腐蚀强度检测模型在各个数据量下的推理时延。The various corrosive substances generated during the corrosion process of hot-dip Zn-Al-Mg coating result in surface images containing a large amount of noise and interference information such as rust spots,scratches,oil stains,etc.Artificial intelligence detection methods mainly based on deep learning will be trapped in complex cyclic iterative processes,resulting in low efficiency.In order to improve the accuracy and efficiency of corrosion strength detection of hot-dip Zn-Al-Mg coatings and reduce computational resource consumption,a lightweight deep learning-based corrosion strength detection method for hot-dip Zn-Al-Mg coatings is proposed.Adopting the lifting wavelet technique to design an adaptive lifting mechanism,utilizing the multi-scale analysis and denoising capabilities of lifting wavelets to preprocess corroded steel plate images and extract clearer corrosion features.Embedding the adaptive enhancement mechanism into the initial level of the convolutional neural network,a convolutional neural network ALS coating corrosion strength detection model is constructed.In order to further reduce the complexity of detection,a model lightweighting method under time delay constraints is used to lightweight the constructed model.By solving the lightweight model,the corrosion strength detection of hot-dip Zn-Al-Mg coating is achieved.The test results show that the design method is accurate for detecting the strength of small and minor corrosion situations,significantly reducing the inference delay of the coating corrosion strength detection model after implementing lightweight treatment at various data volumes.

关 键 词：热浸镀Zn-Al-Mg镀层 轻量化深度学习 提升小波技术 腐蚀强度检测 时延约束 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]