Enhanced Real-time Pavement Damage Analysis: Integrating CNN Edge Detection with YOLO for Accurate Classification

As urban road damage issues become increasingly severe, automated road damage detection systems are becoming critically important. This paper proposes a CNN-YOLO-based road damage detection model aimed at enhancing the accuracy of road damage detection. Our model optimizes the identification process for cracks and potholes by integrating image resizing, minimum filtering, edge detection, and deep learning techniques within a multitask learning framework. Initially, we employ a minimum filter for image resizing to improve detection speed and reduce processing time. Subsequently, by converting the images to grayscale and utilizing YOLO for object recognition, the accuracy of detection is significantly enhanced. Edge detection techniques, including the Laplacian filter, are used to extract road damage features, ensuring accurate identification of various damage types, including longitudinal cracks, transverse cracks, network cracks, and potholes. Our proposed method leverages the rapid scanning capability of the YOLO model and the feature learning ability of CNN to achieve efficient damage detection.Experimental results show that the model achieves over 90% accuracy across different datasets, with outstanding performance on P-SNR and SSI-M metrics, reaching 50.2 dB and 0.84, respectively. This indicates that the proposed CNN-YOLO model offers high accuracy and reliability in detecting road damage. Additionally, the model demonstrates fast computation speed during training and adapts well to various data conditions and damage types.