Automatic Recognition of Pavement Degradation: Case of Rif Chain

Rif Chain (North of Morocco) is subject of ground instabilities. Several studies analyzed landslides’ triggering factors and confirmed that water is a predominant factor. Once a landslide occurs, it damages the roadway paralyzing traffic especially Mediterranean Bypass Road which links the North of Morocco from East to West. To open up population, recovery maintenance requires a heavy budget. Thus, this paper proposes a preventive solution based on automatic recognition of pavement degradations (cracking and tearing) that allow water penetration through pavement until reaching ground support or even deeper layers destabilizing slopes and then triggering landslides. This study applies deep convolutional neural network using pretrained AlexNet model, to automate image classification of road pavement and recognition of three classes (presence of cracking, presence of tearing and absence of degradation). Then to save more time in training and classifying, a second classifier is trained to simply distinguish between two classes (pavement degraded and pavement not degraded).