Data Enhanced Depth Classification Model for Terracotta Warriors Fragments

Layer	Output size	Size of conv kernel	Output channels	Stride
Input layer	256×256		3
conv1	128×128	7×7	64	2
max pool	128×128	3×3	64	2
conv2_x	64×64	$[\begin{array}{l} 3 \times 3 \\ 3 \times 3 \end{array}] \times 2$	$[\begin{array}{l} 64 \\ 64 \end{array}] \times 2$	1
conv3_x	32×32	$[\begin{array}{l} 3 \times 3 \\ 3 \times 3 \end{array}] \times 2$	$[\begin{array}{l} 128 \\ 128 \end{array}] \times 2$	1
conv4_x	16×16	$[\begin{array}{l} 3 \times 3 \\ 3 \times 3 \end{array}] \times 2$	$[\begin{array}{l} 256 \\ 256 \end{array}] \times 2$	1
conv5_x	8×8	$[\begin{array}{l} 3 \times 3 \\ 3 \times 3 \end{array}] \times 2$	$[\begin{array}{l} 512 \\ 512 \end{array}] \times 2$	1
Fc_6d	1×1		6

Table 2. Comparison of effects of traditional classification methods and proposed classification method
View table
Table 2. Comparison of effects of traditional classification methods and proposed classification method
Method Accuracy-Avg /%
SIFT Feature 78.42
Shape Feature 67.55
SIFT+Shape Feature 84.41
Salient geometric Feature 71.32
ResNet18+CBAM+CutMix 88.69

Table 3. Comparison of effects of classification methods before and after optimization of ResNet18
View table
Table 3. Comparison of effects of classification methods before and after optimization of ResNet18
Method Accuracy-Avg /%
ResNet18 84.76
ResNet18+CBAM 86.21
ResNet18+CBAM+CutMix 88.69

Table 4. Comparison of classification results of ResNet18 in CIFAR-10 before and after optimization
View table
Table 4. Comparison of classification results of ResNet18 in CIFAR-10 before and after optimization
Method Accuracy-Avg /%
ResNet18 89.25
ResNet18+CBAM 91.21
ResNet18+CBAM+CutMix 91.73

Tools

Get Citation

Copy Citation Text

Yuehua Yu, Haibo Zhang, Xin Li, Jiaojiao Kou, Kang Li, Guohua Geng, Mingquan Zhou. Data Enhanced Depth Classification Model for Terracotta Warriors Fragments[J]. Laser & Optoelectronics Progress, 2022, 59(18): 1810010

Download Citation

EndNote(RIS)BibTex Plain Text

Set citation alerts for article

Save article for my favorites

Paper Information

Category: Image Processing

Received: Jun. 9, 2021

Accepted: Jul. 28, 2021

Published Online: Aug. 30, 2022

The Author Email: Mingquan Zhou (mqzhou123@126.com)

DOI:10.3788/LOP202259.1810010

Topics

laser devices and laser physics

Lasers and Laser Optics

Laser physics

laser manufacturing

Instrumentation, Measurement and Metrology

Table 1. Network structure and parameters

Table 1. Network structure and parameters

Table 2. Comparison of effects of traditional classification methods and proposed classification method

Table 2. Comparison of effects of traditional classification methods and proposed classification method

Table 3. Comparison of effects of classification methods before and after optimization of ResNet18

Table 3. Comparison of effects of classification methods before and after optimization of ResNet18

Table 4. Comparison of classification results of ResNet18 in CIFAR-10 before and after optimization

Table 4. Comparison of classification results of ResNet18 in CIFAR-10 before and after optimization