Hierarchical Convolutional Neural Network With Feature Preservation and Autotuned Thresholding for Crack Detection

Qiuchen Zhu,Manh Duong Phung,Quang Phuc Ha,Tran Hiep Dinh

doi:10.1109/access.2021.3073921

Abstract

Drone imagery is increasingly used in automated inspection for infrastructure surface defects, especially in hazardous or unreachable environments. In machine vision, the key to crack detection rests with robust and accurate algorithms for image processing. To this end, this paper proposes a deep learning approach using hierarchical convolutional neural networks with feature preservation (HCNNFP) and an intercontrast iterative thresholding algorithm for image binarization. First, a set of branch networks is proposed, wherein the output of previous convolutional blocks is half-sizedly concatenated to the current ones to reduce the obscuration in the down-sampling stage taking into account the overall information loss. Next, to extract the feature map generated from the enhanced HCNN, a binary contrast-based autotuned thresholding (CBAT) approach is developed at the post-processing step, where patterns of interest are clustered within the probability map of the identified features. The proposed technique is then applied to identify surface cracks on the surface of roads, bridges or pavements. An extensive comparison with existing techniques is conducted on various datasets and subject to a number of evaluation criteria including the average F-measure ( <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$AF_\beta $ </tex-math></inline-formula> ) introduced here for dynamic quantification of the performance. Experiments on crack images, including those captured by unmanned aerial vehicles inspecting a monorail bridge. The proposed technique outperforms the existing methods on various tested datasets especially for GAPs dataset with an increase of about 1.4% in terms of <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$AF_\beta $ </tex-math></inline-formula> while the mean percentage error drops by 2.2%. Such performance demonstrates the merits of the proposed HCNNFP architecture for surface defect inspection.

Highlights

Surface inspection plays an important role in the health surveillance and hazard control of roads, bridges, pavements or tunnels
A comparison of Deepcrack architecture and our proposed one is shown in Fig. 2, wherein learning performance can be significantly improved from resized concatenations in FPB so as not to increase the computational latency
COMPARISON WITH RESULTS FROM DIFFERENT FRAMEWORKS The visual results of those deep convolutional neural networks (DCNN) frameworks for crack detection are depicted in Fig. 5, wherein autoencoder models such as SegNet, FPBHN, DeepCrack, and hierarchical convolutional neural network with feature preservation (HCNNFP) are capable of resisting the interference caused by texture, painted boundaries, and uneven lighting conditions

Summary

Introduction

Surface inspection plays an important role in the health surveillance and hazard control of roads, bridges, pavements or tunnels. Effective maintenance and damage prevention of transport infrastructure rely on prompt detection for defects in transportation infrastructure such as cracks, edge failures, potholes, rutting, subsidence, or any surface deterioration [1]. The inspection conducted manually by professional practitioners, wherein dangerous and unattainable sites would limit the effectiveness of human inspection. Successful identification of defect features on infrastructure surface requires the development of feasible, robust and effective detection algorithms. In visual inspection from captured images, an intensity shift indicates a contrast between defective spots and their surrounding pixels in the color space. Thresholding approaches are employed to execute fast detection by solely exploiting the statistics of intensity. Early trials for surface imperfectness detection have been conducted with hybrid utilization of intensity and geometrics [3]

Methods

Results

Conclusion