Hybrid Deep Learning Framework for Road Surface Classification: Integrating Autoencoder-Based Denoising and CNN-Based Classification

This study uses a hybrid deep learning technique to classify asphalt, pavement, and unpaved roads. In real-world circumstances, image data noise can damage image categorization algorithms. This issue can be addressed by a deep neural network (DNN)-based classification system that uses advanced denoising algorithms to improve input images before categorization. We start by denoising noisy native images with autoencoder (AE) approaches. We use two autoencoders: Denoising Autoencoder(DAE) and Convolutional Denoising Autoencoder(CDAE). Proper categorization requires models that filter noise and increase visual clarity. The CDAE employs convolutional layers to maintain spatial hierarchies and local characteristics during denoising, whereas the DAE involves encoding and decoding to rebuild images. The rebuilt images are classified using a CNN after denoising. The CNN is a preferred DNN architecture for this job since it can gather and represent complex visual input. CNN identifies classification-boosting features using noise-free image training. Experiments show this hybrid model works. With 97.92% classification accuracy, the CDAE-CNN architecture could recognize road surface types and conditions under noisy environments. This performance proves the hybrid approach's durability despite training on noise-corrupted images. It improves image classification in noisy data. Denoising algorithms improve deep learning classifier accuracy and make them more relevant in real-world applications with low image quality. These hybrid DAE-CNN/CDAE-CNN models minimize noise and properly categorize road surfaces.