Pipeline Defect Recognition Algorithm Based on CNN Transformer Model for Internal Detection in PE Gas Pipelines

Polyethylene (PE) gas pipelines are integral to the natural gas transportation infrastructure due to their flexibility and resistance to corrosion. However, maintaining their integrity is crucial for preventing gas leaks and ensuring safety. Traditional inspection methods, including manual checks and non-destructive testing techniques, are labor-intensive and may not detect subtle defects. This paper introduces a novel approach leveraging deep learning and the CNN - Transformer model for automated defect recognition in PE pipelines. The model integrates convolutional neural networks for localized feature extraction with a Transformer module to capture global dependencies, enhanced by a Squeeze-and-Excitation module for feature recalibration. Utilizing a dataset of 1200 images representing various defect types, the model demonstrated an average accuracy of 97.8% in defect classification tasks. Experiments showed the model's superiority over traditional CNNs and the Transformer model on medium and small-scale datasets. The findings suggest that the proposed deep learning model offers a reliable, efficient, and accurate method for defect detection in PE pipelines, with potential for real-world application in maintenance and safety management.