A Method for 3D Object Point Cloud Recognition Based on Subflow Pattern Sparse Convolution

Abstract

The invention discloses a method for recognizing a three-dimensional target point cloud based on subflow type sparse convolution, which comprises the following steps: S1 acquires the initial point cloud of the target scene; S2 acquires the local features of the target point cloud; S3 convolves the subflow type The output of the network serves as the input of two identical MLP networks; S4 outputs the first MLP network to the cross-entropy loss function; S5 outputs the second MLP network to two attention-based graph convolutional neural networks, and finally Output to the square error loss function; in addition, the second MLP network is directly output to the square error loss function without the graph convolutional neural network; S6 selects the sum of the cross entropy loss function and the square error loss function as the total loss function of the network model, according to The size of the total loss function value is used for reverse training of the network model. The invention speeds up the network training speed, improves the recognition accuracy, improves the defect of occupying a large memory space, and realizes fast and efficient three-dimensional target recognition.

Description

technical field [0001] The invention relates to the fields of deep learning and three-dimensional target detection and recognition, in particular to a method for identifying three-dimensional target point clouds based on sub-flow type sparse convolution. Background technique [0002] In recent years, convolutional neural networks have set off an upsurge in deep learning, computer vision research and their applications. Because of their powerful feature learning capabilities, they have attracted widespread attention from experts and scholars at home and abroad. However, when using convolutional networks to process some naturally sparse input spatiotemporal data, such as point clouds obtained by lidar scanners or RGB-D cameras, they are inherently sparse. Applying convolutional networks to such sparse data is very inefficient, so how to deal with spatially sparse data more efficiently and use them to develop spatially sparse convolutional neural networks is the top priority of...

AI Technical Summary

Problems solved by technology

In terms of convolution, this is the same as traditional regular convolutional neural networks, but they require less computing resources in terms of floating point operations and memory

Images