Classification of Pavements into Bleeding and Non-Bleeding using Deep Architectures: A Comparative Study

Bleeding or flushing refers to a reflective, dark layer of asphalt on the road caused by the upward movement of asphalt within the pavement. This issue is often caused by excess asphalt in the asphalt concrete, high temperatures, low air void content, and poor asphalt quality. The presence of bleeding poses safety risks as it creates an excessively smooth surface, lacking the necessary texture to prevent hydroplaning. This study presents a comparative analysis of various image classification models for the detection of bleeding and non-bleeding roads. The objective is to develop an automated system that accurately identifies road conditions related to bleeding, which is crucial for effective road maintenance and safety measures. In this regard, four pre-trained deep learning models such as VGG16, InceptionV3, EfficientNet, and ResNet50, to classify road images. The advantages of this work lie in the exploration of multiple state-of-the-art models, allowing for a comprehensive evaluation of their performance. The features and approach of the proposed work involve training the models on a labeled dataset of bleeding and non-bleeding road images. Each model is fine-tuned using transfer learning techniques, leveraging the pre-trained weights from large-scale image datasets. This study highlights several notable contributions about pre-trained deep learning models. In addition, the study identifies the strengths and weaknesses of each model, helping to guide future research and development efforts. The findings of this comparative study contribute to the field of road condition analysis and maintenance by providing valuable insights into the performance and limitations of different image classification models. The better performance is identified with InceptionV3 with an accuracy of 70.00% and least performance is given by ResNet50 with 31.66%.