A Novel Approach for Shadow Removel and Pavement Crack Detection Using Deep Learning

The difficult challenge at hand is automatic pavement crack identification, which involves preserving pavement stability by eliminating shadows that impede fracture detection performance and have the same intensity as cracks. To address the interference caused by shadows, effective algorithm models and training datasets are still lacking. Methodology involves three steps first preprocessing steps improves quality of the image. Second step involves detection of both shadows and removal of shadows whereas in step three classification of cracks is done by using deep learning approach. The system uses intensity values for precision and support multiple image formats. Advanced filtering eliminates noise. Binarizing and filling holes improves crack detection by using cany edge detection. To isolate important features, insignificant blobs are removed and cracks are identified by using feature extraction network. We suggest a data augmentation technique based on this process, which takes into account the variations in brightness brought about by weather and seasonal variations. In the end, we implemented a residual feature augmentation to find tiny fissures that might foretell unexpected catastrophes, and the approach enhances the model's overall performance. Finally, we introduced an deep learning method for classification of cracks that can predict sudden disasters. The system detects cracks at 94% with least MSE. The testing and training accuracy of the proposed model out performs the other state of art methods.