Automatic Defect Description of Railway Track Line Image Based on Dense Captioning

Area of Science:

• Computer Vision
• Artificial Intelligence
• Railway Engineering

Background:

• Railway track line state monitoring is crucial for transportation safety.
• Accurate defect recognition reports are vital for maintenance decisions.
• Automatic generation of inspection reports remains an underexplored area.

Purpose of the Study:

• To develop an automated system for generating railway track inspection reports.
• To leverage dense captioning techniques for describing track components and defects.
• To improve the accuracy and efficiency of railway safety state assessment.

Main Methods:

• Proposed RTLCap model based on DenseCap, utilizing ResNet-50-FPN for feature extraction.
• Incorporated Soft-NMS and Focal Loss to address object occlusion and category imbalance.
• Developed Faster RTLCap using YOLOv3 for improved speed and reduced complexity, featuring MFLMF and SPP in the encoder and stacked LSTM in the decoder.

Main Results:

• RTLCap demonstrated enhanced defect description performance.
• Faster RTLCap achieved improved image processing speed and model efficiency.
• Both quantitative and qualitative experiments validated the effectiveness of the proposed models.

Conclusions:

• The developed RTLCap and Faster RTLCap models offer a promising solution for automated railway track inspection report generation.
• These AI-driven approaches can significantly contribute to ensuring railway transportation safety.
• The methods show potential for real-world application in railway maintenance and monitoring.