Railway Foreign Object Detection Using Normal-Image Anomaly Maps
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional machine vision systems for railway foreign object detection face challenges due to limited accessibility of field data, scarcity of anomalous images, and imprecise anomaly detection, particularly in terms of pixel-wise annotations, leading to unreliable and inefficient detection.
Innovation Solution
A computer-implemented system utilizing a memory-suppress diffusion network and contrastive dissimilarity network trained on normal images to reconstruct and predict anomaly maps, incorporating anomaly rejection mechanisms to enhance robustness and provide both image-level and pixel-wise detection results.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If manual inspection methods are used for railway foreign object detection, then operational simplicity is maintained, but detection precision and efficiency are insufficient
Solution Approach 1:
The patent replaces manual inspection methods with an automated machine vision system that uses deep learning models to detect foreign objects on railway tracks. The system automatically captures images, processes them through neural networks, and identifies anomalies without human intervention, thereby simultaneously improving both detection precision and efficiency.
2Measurement precision
If extensive field data is collected and retained for anomaly detection training, then detection accuracy is improved, but data privacy and intellectual property concerns arise
Solution Approach 1:
The patent extracts and removes sensitive information from training data through anonymization techniques and data processing methods. By extracting only the essential features needed for anomaly detection while removing identifiable information, the system maintains high detection accuracy while protecting data privacy and intellectual property rights.
3Measurement precision
If pixel-wise annotations are obtained through manual processes for anomaly detection training, then detection precision is improved, but time consumption and labor intensity increase
Solution Approach 1:
The patent performs preliminary automated annotation using pre-trained deep learning models to generate pixel-level anomaly masks before final training. This preliminary action provides initial annotations that can be quickly refined rather than created from scratch, significantly reducing the time and labor required for precise pixel-level annotation while maintaining high detection precision.
4Reliability
If conventional machine learning models are used for anomaly detection with limited anomalous samples, then model training is simplified, but detection reliability deteriorates
Solution Approach 1:
The patent inverts the conventional approach by training the model primarily on normal samples rather than anomalous samples. Since anomalous samples are scarce and diverse, the system learns what normal railway tracks should look like and automatically identifies deviations from this normal pattern, thereby improving detection reliability without requiring extensive anomalous training data.
Data Source
AI summary
A computer-implemented system for foreign object detection in a scene. The system includes a memory-suppress diffusion network module adapted to reconstruct a reconstructed image from an encoded image, and a contrastive dissimilarity network adapted to combine the input image and the reconstructed image to predict an anomaly map for the input image. The encoded image is based on an input image, and the memory-suppress diffusion network module and the contrastive dissimilarity network are trained using only normal, real images. The system leverages only normal images in training and does not compromise the detection performance at the inference stage.