基于深度体素卷积神经网络的三维模型识别分类 
下载： 2743次
Recognition and Classification for Three-Dimensional Model Based on Deep Voxel Convolution Neural Network
1 兰州交通大学电子与信息工程学院, 甘肃 兰州 730070
2 兰州交通大学自动化与电气工程学院, 甘肃 兰州 730070
图 & 表
图 1. 3D网格模型的体素化。(a)渲染后的模型;(b)网格模型;(c)体素模型
Fig. 1. Voxelization of 3D mesh models. (a) Rendered models; (b) mesh models; (c) voxelization models
下载图片 查看原文
图 2. 3D模型的旋转数据扩充。(a)马桶模型;(b)椅子模型
Fig. 2. Data expansion of 3D models by rotation transformation. (a) Toilet models; (b) chair models
下载图片 查看原文
图 3. 卷积操作。(a) 2D; (b) 3D
Fig. 3. Convolution operations. (a) 2D; (b) 3D
下载图片 查看原文
图 4. CNN结构
Fig. 4. Structure of convolutional neural network
下载图片 查看原文
图 5. 3D模型识别分类示意图
Fig. 5. Schematic of 3D model recognition and classification
下载图片 查看原文
图 6. 3D模型识别分类结果
Fig. 6. Recognition and classification for 3D models
下载图片 查看原文
表 1不同旋转角度扩充数据集中3D模型识别分类的准确率
Table1. Accuracy rate for recognition and classification of 3D models in expanded dataset with different rotation angles
| Rotation angle /(°) | Accuracy rate /% |
|---|
| 0 | 70.6 | | 120 | 77.1 | | 60 | 83.5 | | 40 | 87.1 | | 30 | 87.7 |
|
查看原文
表 2不同尺寸卷积核下3D模型识别分类的准确率
Table2. Accuracy rate for recognition and classification of 3D models obtained at different sizes of convolution kernel
| Kernel size | Accuracy rate /% |
|---|
| 5×5×5 | 84.3 | | 3×3×3 | 87.7 |
|
查看原文
表 3不同分辨率下3D模型识别分类的准确率
Table3. Accuracy rate for recognition and classification of 3D models obtained at different resolutions
| Resolution | Recognition accuracy rate /% |
|---|
| 24×24×24 | 81.1 | | 32×32×32 | 87.7 |
|
查看原文
表 4不同算法下3D模型识别分类的准确率
Table4. Accuracy rate for recognition and classification of 3D models obtained with different algorithms
| Algorithm | Recognition accuracy rate /% |
|---|
| SPH[15] | 68.2 | | LFD[16] | 75.5 | | 3D ShapeNets[7] | 77.3 | | Proposed algorithm | 87.7 |
|
查看原文
杨军, 王顺, 周鹏. 基于深度体素卷积神经网络的三维模型识别分类[J]. 光学学报, 2019, 39(4): 0415007. Jun Yang, Shun Wang, Peng Zhou. Recognition and Classification for Three-Dimensional Model Based on Deep Voxel Convolution Neural Network[J]. Acta Optica Sinica, 2019, 39(4): 0415007.
PDF全文
