Deep convolutional neural networks for detection of rail surface defects
read more
Citations
Deep Learning for Anomaly Detection: A Survey.
Segmentation-based deep-learning approach for surface-defect detection
Deep Architecture for High-Speed Railway Insulator Surface Defect Detection: Denoising Autoencoder With Multitask Learning
Remaining useful lifetime prediction via deep domain adaptation
A CNN-Based Defect Inspection Method for Catenary Split Pins in High-Speed Railway
References
ImageNet Classification with Deep Convolutional Neural Networks
Very Deep Convolutional Networks for Large-Scale Image Recognition
Very Deep Convolutional Networks for Large-Scale Image Recognition
Histograms of oriented gradients for human detection
Backpropagation applied to handwritten zip code recognition
Related Papers (5)
Frequently Asked Questions (14)
Q2. What future works have the authors mentioned in the paper "Delft university of technology deep convolutional neural networks for detection of rail surface defects" ?
With this in mind, exploring a deep learning approach that would be general enough to be used for automatic detection of other types of rail defects is their immediate future work of interest. In particular, the authors will explore the use of auto-encoders and other deep networks for this purpose.
Q3. What is the latest approach for detection of rail defects?
Unlike signal processing, the use of image processing techniques and image data analysis is a very recent approach for detection of rail defects.
Q4. What are the common choices of features in the detection of rail defects?
Classically in detection from visual data, gradient-based features such as the histogram of oriented gradients (HoG), scale-invariant feature transforms (SIFT), spacial pyramids, and basis functions representations such as Gabor filters are among the common choices of features (see e.g. [20]).
Q5. How is the convolutional neural network trained?
After three convolutional and max-pooling layers, the high-level reasoning in the convolutional neural network is performed via fully-connected layers.
Q6. How can the rail defect classes be classified?
With the proposed small, medium, and large DCNNs, the rail defect classes can be successfully classified with almost 92% accuracy.
Q7. What are the main topics of this paper?
While [21], [22] have focused on identification of track components such as ballast, concrete, wood, and fastener, in this paper the authors focus on the detection and classification of different defects that occur at the rail surface.
Q8. In what fields have different object recognition techniques been used for detecting defects?
In [4], [15]–[17], some signal processing techniques such as noise reduction and wavelet transforms have been used for estimating irregularities and detecting defects in railway tracks.
Q9. What is the classification of normal samples?
In order to convert the multi-class classification results to the binary classification of normal samples versus anomalies, the authors simply regard all the non-normal classes as one and compute the numbers of true positives (TP ), true negatives (TN ), false positives (FP ), and false negatives (FN ).
Q10. What is the history of deep neural nets?
Deep convolutional neural nets have been developed rapidly in the field of object recognition since the breakthrough work of [1].
Q11. How many classes of squats are included in the data?
Among the collected frames, the authors manually labelled 22408 objects as belonging to 1 out of 6 classes (normal, weld, light squat, moderate squat, severe squat, and joint).
Q12. What is the description of the large DCNN model?
From the performance results of the DCNN models, the authors conclude that the large DCNN model performs better for the classification task than the small and medium DCNN model, although the network training takes a longer time.
Q13. What is the weight parameter for the gradient descent?
The weight parameters w are therefore obtained through optimization of the approximated expected value of an error function f defined as:Et[f(w)] = 1b tb∑ i=(t−1)b+1 f(w;xi) (1)where t ∈ {1, ...T} is the iteration index and xi is the ith training sample.
Q14. What is the function that facilitates the discrimination between image classes?
This function facilitates the discrimination between image classes by being imposed on the convolution filter output and performing a nonlinear transformation of a data space.