Deep CNN Approach with Visual Features for Real-Time Pavement Crack Detection

This research delves into an innovative approach to an age-old urban maintenance challenge: the timely and accurate detection of pavement cracks, a key issue linked to public safety and fiscal efficiency. Harnessing the power of Deep Convolutional Neural Networks (DCNNs), the study introduces a cutting-edge model, meticulously optimized for the nuanced task of identifying fissures in diverse pavement types, under various lighting and environmental conditions. Traditional methodologies often stumble in this regard, plagued by issues of low accuracy and high false-positive rates, predominantly due to their inability to adeptly handle the intricate variations in images caused by shadows, traffic, or debris. This paper propounds a robust algorithm that trains the model using a rich library of images, capturing an array of crack types, from hairline fractures to gaping crevices, thus imbuing the system with an astute 'understanding' of target anomalies. One salient breakthrough detailed is the model's capacity for 'context-aware' analysis, allowing for a more adaptive, precision-driven scrutiny that significantly mitigates the issue of over-generalization common in less sophisticated systems. Furthermore, the research breaks ground by integrating a novel feedback mechanism, enabling the DCNN to learn dynamically from misclassifications in an iterative refinement process, markedly enhancing detection reliability over time. The findings underscore not only improved accuracy but also heightened processing speeds, promising substantial implications for scalable real-world application and establishing a significant leap forward in predictive urban infrastructure maintenance.

@article{Kulambayev2024,
  title     = {Deep CNN Approach with Visual Features for Real-Time Pavement Crack Detection},
  journal   = {International Journal of Advanced Computer Science and Applications},
  volume    = {15},
  number    = {3},
  year      = {2024},
  publisher = {The Science and Information Organization},
  author    = {Bakhytzhan Kulambayev and Gulnar Astaubayeva and Gulnara Tleuberdiyeva and Janna Alimkulova and Gulzhan Nussupbekova and Olga Kisseleva},
  doi       = {10.14569/IJACSA.2024.0150333},
  url       = {https://doi.org/10.14569/IJACSA.2024.0150333}
}