Concatenated Residual Attention U-Net with DenseNet: A Hybrid Deep Learning Framework for Pavement Condition Evaluation and Maintenance Planning

Recent days, pavement management systems are important for country's economic well-being as road authorities strive to protect their valuable road assets and minimize the maintenance costs which optimize resource allocation. Traditional approaches for accurate pavement condition evaluation faced significant challenges which include infrequent evaluations lead to delayed maintenance by causing further deterioration, insufficient funding lead to inadequate maintenance. Therefore, this paper proposes a Deep Learning (DL) based Concatenated Residual Attention U-Net with DenseNet (CRAUNet-DenseNet) for Accurate Pavement Condition Evaluation and Maintenance Planning. Initially, data is collected from pavement surface images by using two images. Then, preprocessing is done by using data augmentation to improve training process. After that, image segmentation is done by CRAUNet where, it has encoder, decoder, and skip connection parts to make image segments as well as restore feature maps to its original size. The image features were extracted by utilizing Spatial Pyramid Pooling Layer (SPPL) which allows arbitrary scales of any sizes of image. At last, DenseN et is used for determining and classify condition of road from pavement surface images. The proposed CRAUNet-DenseN et model generated outperforming results in precision 97.91 %, recall 98.36%, and F1 score 99.10% when compared with existing Convolutional Neural Network (CNN).