An effective detection and classification of road damages using hybrid deep learning framework

The monitoring of road surfaces is a critical thing in transport infrastructure management. The manual reporting process increases the processing delay and causes challenges in accuracy. Detecting road surface damages is important for improving the quality of transportation and avoiding several issues normal people face in daily life. Therefore, an automated monitoring system is needed to compute road surface conditions for effective road maintenance regularly. Accurately detecting and classifying road damage images become a challenging task for researchers. Thus, the proposed work introduced a hybrid deep learning framework for detecting and classifying road damage images. At first, the input images are acquired from the dataset and pre-processed with an adaptive intensity limited histogram equalization algorithm. This pre-processing method enhances the contrast of the given input images and eliminates the noise presented in the image. Then, the damage detection is performed in the segmentation stage using an adaptive density based fuzzy c-means clustering method. Features from the segmented images are extracted using Laplacian edge detection with Gaussian operator and hybrid wavelet-Walsh transform approaches. Subsequently, the dimensionality of the feature set is reduced by using the Adaptive Horse herd Optimization (AHO) algorithm. Finally, the road damages are detected and classified using the proposed Hybrid Deep Capsule autoencoder based Convolutional Neural network (Hybrid DCACN) with Improved Whale Optimization (IWO) model. The experimental validation is done using the RDD2020 dataset, and the performance metrics are evaluated to show the efficacy of the proposed model. The proposed work attains 98.815% accuracy, and the obtained results outperform the existing approaches.