| Summary: | Detecting cracks on concrete surfaces is a crucial task in civil engineering inspections, but it poses significant challenges due to the small and concealed nature of cracks. Visual detection is particularly difficult on uneven or rough concrete surfaces. To overcome these challenges, our research focuses on developing an automated system that utilizes a wall-climbing robot for crack classification. Our main objective is to introduce a crack classification technique using MobileNetV2, enabling real-time classification without human intervention. The Convolution Neural Network (CNN) model used for crack classification is based on MobileNetV2, which is fine-tuned by adjusting the sensitivity of its hyperparameters. Through extensive experiments, we evaluate the performance of this CNN approach specifically designed for embedded systems. After evaluating our proposed approach of crack-detection on publicly available datasets, we have found that out of all the pre-trained CNN models MobileNetV2 yields the best performance with 99.56% detection accuracy, precision of 99.65%, recall of 99.48%, and F1-Score of 99.56%. However, it is important to note that the training time for this model is relatively high, taking 25,500 s. Future study of the study should focus on optimizing the computation time to improve efficiency. © 2024, The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd.
|
|---|