Comparison of classic object-detection techniques for automated sewer defect detection

Abstract

Sewer systems play a key role in cities to ensure public assets and safety. Timely detection of defects can effectively alleviate system deterioration. Conventional manual inspection is labor-intensive, error-prone and expensive. Object detection is a powerful deep learning technique that can complement and/or replace conventional inspection manner, especially in complex environments. This study compares two classic object-detection methods, namely faster region-based convolutional neural network (R-CNN) and You Only Look Once (YOLO), for the detection and localization of five types of sewer defects. Model performances are evaluated based on their detection accuracy and processing speed under parameterization impacts of dataset size and training parameters. Results show that faster R-CNN achieved higher prediction accuracy. Training dataset size and maximum number of epochs (MaxE) had dominant impacts on model performances of faster R-CNN and YOLO, respectively. The processing speed increased along with the increasing training data for faster R-CNN, nevertheless, did not vary significantly for YOLO. The model abilities in detecting disjoint and residential wall were highest, whereas crack and tree root were more difficult to detect. The results help to better understand the strengths and weaknesses of the classic methods and provide a user useful guidance for practical applications in automated sewer defect detection.

Automatic Detection Method of Sewer Pipe Defects Using Deep Learning Techniques

“…The emergence of deep learning techniques has better addressed the drawbacks of traditional computer vision methods. As the most popular algorithm among them, convolutional neural network (CNN) is most widely used [10]. Compared with the traditional CV method, CNN can automatically extract image features and perform recognition, without the need for professional technicians with rich work experience and complex feature design process, which greatly simplifies the detection process.…”

“…The method demonstrates the high accuracy of CNN in sewer defect classification tasks and has been applied to practical inspection tasks. However, this method can only detect one type of defect on a single image, whereas sewer pipes may have multiple types of defects at the same location [10].…”

“…The sewer defects studied in this paper are four types commonly encountered in southern China [10,21], namely crack, deposition, root and stagger. The original images were obtained from the Danish laboratory's public dataset Sewer-ML [22].…”

A Composite Transformer-Based Multi-Stage Defect Detection Architecture for Sewer Pipes

“…On the other hand, object detection architectures are mainly classified into anchorfree models [14], single-stage models [15], two-stage models [2], and multi-stage models [16]. It has been shown that two-stage is more likely to achieve superior precision than single-stage when there is more training data on sewer defects [17]. Inspired by previous work, a novel multi-stage object detection model based on a composite backbone Swin Transformer is designed for detecting sewer defects.…”

A Lightweight Method for Detecting Sewer Defects Based on Improved YOLOv5

“…With the development of deep learning technology, object detection technology that can perform both classification and detection is widely used for vehicle recognition [4], hand script counterfeit detection [5], and automatic detection of drainage sewers in China and abroad [6,7]. The most commonly used methods include Single Shot Multibox Detector (SSD) [8], You Only Look Once (YOLO) [9], and Faster RCNN [10], which effectively improve the detection accuracy by using neural networks to automatically extract features.…”