New Classification of Drainage Pipe Defects Based on Computer Vision and Database Building and Effect Analysis

Abstract

Identifying and detecting defects in drainage pipes are essential for maintaining drainage pipe networks. This process begins by building a database for defect classification. Currently, the “Technical regulations for detection and evaluation of urban drainage pipes” is the main reference for the classification of drainage pipe defects and the building of the database. However, the regulations do not consider computer vision requirements in the classification system. This research suggests a novel computer vision-based classification scheme for drainage pipe flaws that is divided from the initial two categories of 16 defects into three categories of 12 defects. A database was constructed using engineering samples to focus on six of the most harmful defects out of the 12 categories identified categories: rupture; interface deviation; surface damage; roots; leakage; and sealing defects. Experiments were conducted on the created database using the YOLOv8 model to analyze: (1) the impact of varying library sizes on detection performance; (2) the influence of sample libraries in different environments (varying brightness and contrast); (3) augmentation and updating of sample data; and (4) analysis of the impact of sample libraries after data augmentation and updating. The experimental findings demonstrate that the model achieves a precision (P-value) of 71.6%, a recall (R-value) of 63.7%, and a mean accuracy (mAP-value) of 71.4%. This value initially ensures that the new classification system and the database are compatible with the current computer vision technology and can yield favorable outcomes.