Asphalt Pavement Bleeding Evaluation using Deep Learning and Wavelet Transform

Pavement inspection is an important part of pavement management systems because this part provides input and raw material information to the system. If the pavement situation has not been assessed or incorrectly assessed, it will not be possible to carry out optimum maintenance and repair operations. It can also cause higher maintenance costs and the risk of accidents. Pavement distress information is crucial data that should be collected and evaluated in the pavement inspection process. Accordingly, wide research has been conducted to develop more efficient systems for the evaluation of pavement distresses using new technologies. Bleeding is one of the asphalt pavement distresses, which directly affects the skid resistance and vehicle maneuverability. Based on the literature, pavement bleeding received the attention from the research community less than other pavement distress such as cracks, rutting, raveling, and potholes. This research attempts to develop an efficient system for the automatic evaluation of asphalt pavement bleeding. For this aim, the transfer learning method was applied to train a pre-trained convolutional neural network for bleeding detection. Also, various image processing techniques (wavelet transform analysis as the main technique) were used to segment bleeding regions in pavement images. Results indicated that the proposed system has good performance in bleeding detection and segmentation with 98% and 87%, respectively. Accordingly, this system can be applied as an efficient system for pavement bleeding evaluation.