Machine Learning Based Deep Cloud Model to Enhance Robustness and Noise Interference

Abstract

To improve the ability of the 3-D point cloud deep network classification model, this paper presents a competitive attention fusion module that may be applied to various classification networks. The intrinsic similarity between representation and intermediate features is used to redistribute the weights of intermediate feature channels. Implementation of the present module in the standard networks such as 
Point net++ and Point ASNL, were included to conduct the experiments. The suggested block is autonomous and transferrable, according to the results, and it concentrates on the core and backbone elements that are better suited for 3D point cloud form categorization. The presented module improves the model's resistance to point cloud disturbance noise, outlier noise, and random noise when compared to the state-of-the-art network without affecting classification accuracy. When the random noise numbers are 0, 10, 50, 100, and 200, the accuracies reach 93.2%, 92.9%, 85.7%, 78.2%, and 63.5%, respectively.

Keywords: Machine Learning, Deep Learning, Cloud computing, Classification, Cloud disturbance, Anti-interference