Rail surface defect inspection based on machine vision faces challenges against the complex background with interference and severe data imbalance. To meet these challenges, in this article, we regard defect detection as a key-point estimation problem and present the proposed attention neural network for rail surface defect detection via consistency of Intersection-over-Union(IoU)-guided center-point estimation (CCEANN). The CCEANN contains two crucial components. The two components are the stacked attention Hourglass backbone via cross-stage fusion of multiscale features (CSFA-Hourglass) and the CASIoU-guided center-point estimation head module (CASIoU-CEHM). Furthermore, the CASIoU-guided center-point estimation head module integrating the delicate coordinate compensation mechanism regresses detection boxes flexibly to adapt to defects’ large-scale variation, in which the proposed CASIoU loss, a loss regressing the consistency of intersection-over-union (IoU), central-point distance, area ratio, and scale ratio between the targeted defect and the predicted defect, achieves higher regression accuracy than state-of-the-art IoU-based losses. The experiments demonstrate that the CCEANN outperforms competitive deep learning-based methods in four surface defect datasets.

Visual inspection techniques for rail surface defects have become prevalent approaches to obtain information on rail surface damage. However, uneven illumination leads to illegibility of local information, and the change of the wheel-rail area results in the changeful background of the rail surface, both of which pose challenges to the visual inspection. This paper proposes a novel algorithm that detects rail surface defects via partitioned edge features (PEF). PEF eliminates the effect of uneven illumination by effectively extracting edge features and building homogeneous background on the rail surface. In the process of edge feature extraction, the thresholding based on adaptive partition of rail surface (APRS) plays an indispensable role. In APRS, the rail surface is adaptively partitioned into three types of regions according to the wheel-rail contact degree. After that, the dynamic threshold is set adaptively for each region type on the basis of the prior information of defect proportion. Subsequently, based on neighborhood information and fuzzy decision, the spatial information of adjacent pixels and the direction information of fracture edges are utilized to realize the effective recovery of incomplete defect contours. In addition, defect contours are precisely filled via a flexible combination of morphological hole filling operation and defect region extraction based on improved background difference. The accuracy of this PEF algorithm was confirmed by experiments and comparisons with related algorithms. The experiment results show that PEF detects defects with 92.03% recall and 88.49% precision, which achieves higher accuracy than the established detection algorithms for rail surface defects.

Fastener needs to be monitored and inspected periodically to ensure the rail’s safety due to its easily damaged accessory for railway infrastructure. Recently, Industrial Internet of Things (IIoT) and artificial intelligence (AI)-based visual inspection techniques have been exploited to realize the online inspection of fastener’s fault by using a fault detection IoT vehicle that is mounted with multitype sensors and cameras according to the design of our research team. However, instead of traditional artificial inspection, the AI-based automatic fastener inspection approach is still faced with some challenges, for example, collection of enough samples of faulted fastener. In this article, we propose a cascade learning embedded vision inspection method of rail fastener based on the deep convolutional neural network (DCNN). The proposed method has two steps: 1) region position and 2) fault detection. First, a modified single shot multibox detector (SSD) model is adopted to locate the fastener regions from the captured railway images. Then, a key component detection (KCD) method based on the improved faster region convolutional neural network (RCNN) is proposed to realize the detection of faulted fastener. Extensive experiments are conducted to demonstrate the performance of the proposed method. The experiment results show that the proposed method achieves an average precision of 95.38% and an average recall of 98.62% on fastener detection, which is much better than the manual operation.

This article aims to improve the performance of deep learning-based defective fastener inspection method. Due to the defective fasteners are insufficient and far less than the defect-free ones in real railway, it is difficult to train a robust fastener inspection model on such imbalanced dataset. In view of this problem, a novel image generation method called four-discriminator cycle-consistent adversarial network (FD-Cycle-GAN) is proposed to generate the defect fastener images using a large number of defect-free ones. Extensive experiments are conducted on the real fastener images and generated images. Experimental results demonstrate that the defect fastener images generated by our proposed method have better quality and richer diversity than those generated by other state-of-the-art methods. In addition, compared with the CNN-only baseline, the performance of the fastener inspection model trained on the expanded dataset containing the defect fastener images generated by FD-Cycle-GAN is improved significantly. The detection accuracy and relative IMP reach 93.25% and 21.59% respectively.

Fasteners are an important part of railway security. The application of computer vision and image processing on fastener detection has become a research hot spot. However, most of the researches have focused on the missing and damage detection of fasteners up to now. In this paper, a method based on image processing is proposed to locate the center of the hexagon nut in the railway fastener. The precise position of the hexagon nut can be used for the automatic replacement of the nut, which is of great significance for the automatic railway construction and maintenance. The locating method proposed in this paper can be divided into two steps. First, the improved Canny detector which combines surrounding suppression and iterative excitation is used to extract the contours of the fastener. Then, a multifeature hierarchical locating (MHL) algorithm is proposed, which integrates the circle contour of the gasket and the hexagon features of the nut. The experimental results show that the improved Canny detector suppresses noises effectively and highlights the contours of fastener. The MHL algorithm in the next step improves the locating accuracy remarkably while guaranteeing the robustness meantime.

For rail wear measurement, the process of profile registration is critical. Nowadays, the widely used method is the rail waist double circle segment method (DCS). However, this method may be invalid in actual applications because of the influence of the following two factors: one is the outliers mixed into profiles and the other is the profile diversity. The former induces the misalignment between the measured profile and the standard one, and the latter may destroy the integrity of rail waist, which makes that the measured profile cannot be matched with the standard one through DCS. To solve the two problems, a hybrid profile registration method is proposed in this paper. First, by locating new matching primitives to realize coarse registration, we check the profile integrity fast. Then, for complete profile, we rematch its original measured profile with the standard one finely based on DCS to detect and remove the outliers correctly. The efficiency and superiority of the proposed method were verified by numerous experiments. The results show that the average score of F1-Measure for outlier detection reaches 0.95, which outperforms some classical models obviously. Meanwhile, the result of the profile integrity check is also basically coincided with the real face of test zones. Furthermore, the system can run at a speed of 21.95 km/h under our experimental setup, which is far higher than that of the rail maintenance train (up to 5 km/h).