Shape Classification Detection Algorithm Based on Improved MeshNet Network
YANG Junjie
WANG Jun
CHENG Yong
Abstract：With the progress of hardware,convolutional neural network has been rapidly developed in the field of two-dimen-sional image recognition,the promotion of convolutional neural network to three-dimensional has gradually become a new demand and trend.In different three-dimensional representations,because of its complex structure and irregular shape,mesh-based model brings difficulties to the application of grid data in the field of deep learning until the emergence of MeshNet reversed this situation.This paper improves MeshNet,introduces the Self-attention and ResNet,establishes the correlation of input and expands the depth of the network.A new loss function PolyLoss is introduced to replace the original Cross-entropy loss function,which can achieve better performance than Cross-entropy loss function.Finally,this paper evaluates the shape classification task after the improvedal-gorithm.The experimental results show that the accuracy of this method and the average detection accuracy of all categories reach 92.1%and 85.1%,respectively,0.9%and 4.3%higher than the original MeshNet.
Keywords：convolutional neural networkResNetMeshNetself-attentionPolyLoss
Publication Date：2025-09-20
Online Publishing Date：2025-12-23(First online date of this platform, not the publication date of the document)
Pages：6( 2393-2397,2441 )