Characterization and identification of asphalt mixtures based on Convolutional Neural Network methods using X-ray scanning images

Accurate identification of asphalt mixtures based on images is a difficult task due to the complex components of materials under random mixing processes. This study explores the application of a Convolutional Neural Network (CNN) in classifying and identifying asphalt mixtures using the sectional images obtained from the X-ray computed tomography (CT) method. Images of 11 asphalt mixtures with two mixing types (hot mix and cold recycled mix), three nominal maximum aggregate sizes (13, 20 and 25 mm), and two compaction methods (Marshall and Superpave Gyratory) were obtained and segmented into three parts, including air voids, mastics, and aggregates, based on digital image processing (DIP) methods. The CNN models were trained on different combinations of image sets, and data augmentation technique was applied to the train set. Test results showed that CNN models from binary images of mastics and aggregates could success classify different asphalt mixtures with higher reliability than CNN model from images of air voids. CNN model from mastic could better distinguished different compaction methods while CNN model from aggregates could recognize different gradation types better.