A Texture Integrated Deep Neural Network for Semantic Segmentation of Urban Meshes

3-D geo-information is essential for many urban related applications. Point cloud and mesh are two common representations of the surface. Compared to point data, possesses indispensable advantages, such as high-resolution image texture sharp geometry representation. Semantic segmentation, an important way obtain geo-information, however, mainly performed on data. Due complex representation lack efficient utilizing information, semantic segmentation still a challenging task acquisition. In this article, we propose integrated deep learning method task. A novel convolution module introduced capture features. The features concatenate with nontexture that represents by center gravity (COG) triangles. hierarchical network employed segment COG cloud. Our experimental results show proposed significantly improves accuracy (1.9% overall 4.0% average F1 score). It also compares other state-of-the-art methods public SUM-Helsinki dataset achieves considerable results.