Multi-Feature Fusion and Visualization of Pavement Distress Images Based on Manifold Learning

Abstract

For multi-feature fusion in automatic recognition of pavement distress images, we proposed a multi-feature fusion method based on manifold learning. In this method, the intrinsic features of pavement distress images are extracted through mapping the high dimensional data combing projection, mixture density factor and second order moment invariant into the low dimensional space. The multiple features are fused and the visualization of pavement distress images is implemented. In the experiments, we applied the multi-feature fusion method in the detection of pavement distress images. Two-dimensional features are first extracted from the 8 combining features, then the recognition effects on the 2D features of 4 methods including ELM, KNN, SVM and BP network are compared. The experimental results show that the proposed method effectively improved the detection accuracy of pavement distress images. Simultaneously, the physical meaning of the 2D features is obtained through visualizing. One feature preliminary denotes the complexity and damaged extent of the cracks in images, the other describes the direction of the cracks.