Laser & Optoelectronics Progress, Volume. 58, Issue 20, 2010015(2021)
Research on Classification Method of Sand and Gravel Aggregate Based on Convolutional Neural Network
Figures & Tables(11)
Fig. 3. Schematic of sand and gravel aggregate. (a) Sand and gravel aggregate with dry surface; (b) sand and gravel aggregate with wet surface
Fig. 4. Schematic of all grades of sand and gravel aggregate. (a) 1st level; (b) 2nd level; (c) 3rd level; (d) 4th level; (e) 5th level
Table 1. Parameter settings of each network layer in CNN13
View tableTable 1. Parameter settings of each network layer in CNN13
Network layer Input Filter Output Input 384×275×1 384×275×1 conv 1-64 384×275×1 3×3×64 384×275×64 maxpooling 1 384×275×64 2×2 192×138×64 Network layer Input Filter Output conv 2-128 192×138×64 3×3×128 192×138×128 maxpooling 2 192×138×128 2×2 96×69×128 conv 3-256 96×69×128 3×3×256 96×69×256 conv 4-256 96×69×256 3×3×256 96×69×256 maxpooling 3 96×69×256 2×2 48×35×256 conv 5-512 48×35×256 3×3×512 48×35×512 conv 6-512 48×35×512 3×3×512 48×35×512 maxpooling 4 48×35×512 2×2 24×18×512 conv 7-512 24×18×512 3×3×512 24×18×512 conv 8-512 24×18×512 3×3×512 24×18×512 maxpooling 5 24×18×512 2×2 12×9×512 conv 9-512 12×9×512 3×3×512 12×9×512 conv 10-512 12×9×512 3×3×512 12×9×512 maxpooling 6 12×9×512 2×2 6×5×512 FC 1-1024 6×5×512 1024 FC 2-1024 1024 1024 FC 3-1024 1024 1024
Table 2. Classification standard of sand and gravel aggregate
View tableTable 2. Classification standard of sand and gravel aggregate
Grade Particle size /mm Mass fraction /% Needle and flake(Q) Round or square(P) 1 10-15 0≤Q≤2 98<P≤100 2 10-20 2≤Q≤5 95<P≤98 3 10-25 5≤Q≤10 90<P≤95 4 10-25 10≤Q≤20 80<P≤90 5 10-25 20≤Q≤30 70<P≤80
Table 3. Comparison of original image before and after preprocessing
View tableTable 3. Comparison of original image before and after preprocessing
Image Size /MB Width /pixel Height /pixel Original image 10.00 3840 2748 Processed image 0.10 384 275
Table 4. Comparison of memory consumption between CNN13 model and VGG16 model
View tableTable 4. Comparison of memory consumption between CNN13 model and VGG16 model
Model Number of
convolution layersTotal
parametersModel
memory /MBOutput
memory /MBMaximum
batchsizeCNN13 13 31692160 120.9 73.2 37 VGG16 16 138357544 527.8 58.1 33
Table 5. Accuracy of CNN13 model and VGG16 model
View tableTable 5. Accuracy of CNN13 model and VGG16 model
Model Accuracy /% Grade 1 Grade 2 Grade 3 Grade 4 Grade 5 CNN13 100.0 100.0 99.5 99.5 100.0 VGG16 97.5 100.0 99.5 81.0 93.0
Tools
Get Citation
Copy Citation Text
Ran Yan, Jideng Liao, Xiaoyong Wu, Changjiang Xie, Lei Xia. Research on Classification Method of Sand and Gravel Aggregate Based on Convolutional Neural Network[J]. Laser & Optoelectronics Progress, 2021, 58(20): 2010015
Paper Information
Category: Image Processing
Received: Nov. 28, 2020
Accepted: Jan. 11, 2021
Published Online: Oct. 13, 2021
The Author Email: Yan Ran (yanran@cqut.edu.cn)
© Copyright 2018-2021 | Chinese Laser Press.
All Rights Reserved 沪ICP备15018463号-20