Foreign Object Detection in Railway Images Based on an Efficient Two-Stage Convolutional Neural Network.

Area of Science:

• Computer Science
• Artificial Intelligence
• Machine Learning

Background:

• Foreign object intrusion on railways is a significant cause of accidents, posing risks to human life and property.
• Current manual detection methods for railway foreign objects are inefficient and prone to errors.
• Real-time detection of railway intrusions is crucial for preventing train collisions.

Purpose of the Study:

• To propose an efficient two-stage framework for foreign object detection in railway images.
• To develop a lightweight railway image classification network for real-time detection.
• To enhance the accuracy and efficiency of foreign object detection systems.

Main Methods:

• A two-stage framework comprising image classification and object detection.
• A lightweight classification network utilizing improved inverted residual units.
• Incorporation of selective kernel convolution for multiscale feature learning.
• Integration of a convolutional block attention module for enhanced feature representation.

Main Results:

• The classification network achieves performance comparable to widely used baselines with superior efficiency.
• The proposed framework demonstrates effective real-time classification of normal versus intruded railway images.
• The second-stage object detection network provides satisfying performance in locating and classifying foreign objects.

Conclusions:

• The developed two-stage framework offers an efficient and accurate solution for railway foreign object detection.
• The enhanced inverted residual unit with selective kernel convolution and attention mechanisms improves classification performance.
• This approach contributes to enhanced railway safety by enabling timely detection of potential hazards.