Improved GoogLeNet-Based Crack Detection in Concrete Structures

Crack detection in concrete structures is critical for the maintenance and safety of concrete structures. Traditional methods, which often rely on manual inspection, are labour-intensive and subject to human bias. This paper presents an advanced crack detection system leveraging an improved GoogLeNet architecture which is convolutional neural network (CNN) known for its depth and efficiency. The proposed system achieves superior accuracy and robustness in identifying cracks from images by incorporating enhancements such as optimized inception modules, refined hyperparameters, and transfer learning techniques. This work adopted a publicly available CCIC dataset for training and validation purposes. The accuracy of the proposed method is compared with the latest state-of-the-art methods based on the GoogLeNet published in the literature. The obtained training and validation accuracy of the improved GoogLeNet model presented in this paper are 98.6%, and 98.5% respectively. These are the highest results of GoogLeNet for crack detection in concrete structures reported to date. The proposed approach has proven to be highly effective for crack detection in concrete structures and could lead to similar improvements for other types of defect detection.