基于图像增强与深度学习的钢轨表面缺陷检测

Rail surface defect detection based on image enhancement and deep learning

相比传统的物理检测算法,基于机器视觉的检测算法具有检测速度快、操作便捷等诸多优点,但因受光照不均、相机失焦抖动、雨雪天气等外界因素的影响,导致检测精度降低。针对这一问题,提出一种基于图像增强与深度学习的钢轨表面缺陷视觉检测算法。首先,对图像进行Gabor滤波去噪,以减少噪声对缺陷检测的影响;然后,利用HSV空间变换方法增强缺陷图像的关键特征信息;最后,通过改进Faster R-CNN卷积神经网络,实现了多尺度钢轨表面缺陷的检测与识别。通过对所提出的检测算法进行对比实验,实验结果表明:裂纹、剥落、磨损三类缺陷的识别精度分别为91.87%,92.75%和91.52%,检测速度为每张图像0.265 s,优于已有的钢轨表面缺陷检测算法,能够很好地应用于实际项目中。

Compared with traditional physical detection methods,computer-vision-based detection methods has many advantages such as its fast detection speed and convenient characteristics.However,due to the influence of external factors such as uneven illumination,out-of-focus of camera jitter,rain and snow weather,the detection accuracy was reduced.To solve this problem,this paper presents a visual detection algorithm for rail surface defects based on image enhancement and deep learning.Firstly,Gabor filtering was carried out to reduce the impact of noise on the defect detection effect.Then,the key feature information in the image was enhanced by HSV space transformation.Finally,Faster R-CNN convolutional neural network was improved to realize the detection and recognition of multiscale rail surface defects.The proposed algorithm was compared in detail.The experimental results indicate that the proposed algorithm can achieve high accuracy of crack,spalling and abrasion with 91.87%,92.75%and 91.52%,high detect speed with 0.265s per image,substantially outperforming the state-of-the-art rail surface defect detection algorithms.The proposed method can be used for actual fault detection of freight train images.