1. A GRAPH CONVOLUTIONAL NEURAL
NETWORK FOR CLASSIFICATION OF
BUILDING PATTERNS USING SPATIAL VECTOR
DATA
2. INTRODUCTION
• Machine learning methods, specifically, convolutional neural networks(CNNs),
have merged as an integral part of scientific research in many disciplines.
• However, these power full methods of ten fail to perform pattern analysis and
knowledge mining with spatial vector data because in most cases, such data are
not underlying grid-like or array structures but can only be modelled as graph
structures.
3. APPLICATION OF CONVOLUTIONAL NEURAL
NETWORKS TO GRAPHS
• CNN soften fail to analyze spatial vector data because of their regularity
requirements for data structures. To generalize CNNs to general structures such
as graphs, two potential solutions are proposed
• The spatial approach and
• The spectral approach.
5. CONCLUSIONS AND FUTURE OUTCOMES
• Conclusions and outlook Previously, spatial vector data were not sufficient for
performing patterns analysis and knowledge mining when using powerful
convolutional networks because in most cases, the data can only be modeled as a
graph structure that does not satisfy the requirements for regularity.
• To over come this problem, the present study introduced the graph
convolutional neural network(GCNN), where by the convolution operation is
converted from the vertex domain into a point-wise product in the Fourier
domain.