The invention provides a steel rail surface defect detection method and device based on few-sample learning, and the method comprises the steps: carrying out the number expansion of a small number ofmarked steel rail surface defect images, and constructing a steel rail surface defect data set; constructing a steel rail surface defect detection network model with multi-level feature fusion, wherein a multi-task loss function of the steel rail surface defect detection network model with multi-level feature fusion comprises a target positioning bounding box loss function and a multi-class classification loss function, the target positioning bounding box loss function is used for training the model to position a defect area, and the multi-class classification loss function is used for training a model to identify defect classes; training a model according to the steel rail surface defect data set; and after a plurality of steel rail images are collected, obtaining the spliced steel rail images, inputting the spliced steel rail images into the steel rail surface defect detection network model with multi-level feature fusion, and outputting a steel rail surface defect detection result.According to the method and device, the steel rail surface defect detection network model can be trained under a small number of labeled samples, and the effect is good.

    本发明提供了一种少样本学习的钢轨表面缺陷检测方法及装置,该方法包括:对少量已标注的钢轨表面缺陷图像进行数量扩展,构建钢轨表面缺陷数据集;构建多层级特征融合的钢轨表面缺陷检测网络模型,多层级特征融合的钢轨表面缺陷检测网络模型的多任务损失函数包括目标定位边界框损失函数和多类别分类损失函数,目标定位边界框损失函数用于训练模型定位缺陷区域,多类别分类损失函数用于训练模型识别缺陷类别;根据钢轨表面缺陷数据集训练模型;在采集到多张钢轨图像后,获得拼接后的钢轨图像并输入至多层级特征融合的钢轨表面缺陷检测网络模型中,输出钢轨表面缺陷检测结果。本发明可以在少量已标注的样本下训练钢轨表面缺陷检测网络模型,效果好。


    Zugriff

    Download


    Exportieren, teilen und zitieren



    Titel :

    Steel rail surface defect detection method and device based on few-sample learning


    Weitere Titelangaben:

    少样本学习的钢轨表面缺陷检测方法及装置


    Beteiligte:
    LIU JUNBO (Autor:in) / WANG SHENGCHUN (Autor:in) / WANG FAN (Autor:in) / DU XINYU (Autor:in) / GU ZICHEN (Autor:in)

    Erscheinungsdatum :

    2021-02-05


    Medientyp :

    Patent


    Format :

    Elektronische Ressource


    Sprache :

    Chinesisch


    Klassifikation :

    IPC:    G06T Bilddatenverarbeitung oder Bilddatenerzeugung allgemein , IMAGE DATA PROCESSING OR GENERATION, IN GENERAL / B61K Andere Hilfseinrichtungen für Eisenbahnen , OTHER AUXILIARY EQUIPMENT FOR RAILWAYS / G06K Erkennen von Daten , RECOGNITION OF DATA / G06N COMPUTER SYSTEMS BASED ON SPECIFIC COMPUTATIONAL MODELS , Rechnersysteme, basierend auf spezifischen Rechenmodellen



    Information collection unmanned aerial vehicle for steel rail surface defect detection based on deep learning network

    YONESO MAKOTO / CONG CHAO / JU BENXIANG et al. | Europäisches Patentamt | 2022

    Freier Zugriff

    Vehicle-mounted steel rail defect detection method

    ZHANG HUI / XU CHEN / FAN GUOPENG et al. | Europäisches Patentamt | 2022

    Freier Zugriff

    Research on deep learning method for rail surface defect detection

    Feng, Jiang Hua / Yuan, Hao / Hu, Yun Qing et al. | IET | 2020

    Freier Zugriff

    Steel rail defect detection method based on phased array ultrasound

    WENG CHENXIN / KIM HO-YOUN / WU ERYONG et al. | Europäisches Patentamt | 2023

    Freier Zugriff

    Rail defect detection using ultrasonic surface waves

    Edwards, R.S. / Jian, X. / Fan, Y. et al. | Tema Archiv | 2006