Coal and gangue identification method based on the intensity image of lidar and DenseNet

Jichuan Xing; Zimo Zhao; Yaozhi Wang; Liang Nie; Xian Du

doi:10.1364/AO.422498

Applied Optics
Vol. 60,
Issue 22,
pp. 6566-6572
(2021)
•https://doi.org/10.1364/AO.422498

Coal and gangue identification method based on the intensity image of lidar and DenseNet

Jichuan Xing, Zimo Zhao, Yaozhi Wang, Liang Nie, and Xian Du

Not Accessible

Your library or personal account may give you access

Get PDF
Email
Share
Get Citation
Copy Citation Text
Jichuan Xing, Zimo Zhao, Yaozhi Wang, Liang Nie, and Xian Du, "Coal and gangue identification method based on the intensity image of lidar and DenseNet," Appl. Opt. 60, 6566-6572 (2021)

Export Citation
- BibTex
- Endnote (RIS)
- HTML
- Plain Text
Citation alert
Save article

Check for updates

Abstract

Coal and gangue (rock) identification is the essential process in a coal preparation plant. In an actual coal preparation plant, the existing classification methods have many disadvantages in safety and identification rate. We utilized the echo intensity image (EII) of lidar for coal and gangue identification for the first time, to the best of our knowledge, and achieved outstanding recognition results with a convolutional neural network. First, we acquire the information of the 3D point cloud, including the distance and the echo intensity, and decompose them into two channels. Then, we utilize the distance channel to remove the background noises and separate the object and the echo intensity channel to construct the 2D EII. Finally, we prune the dense convolutional network (DenseNet-121) to DenseNet-40 for the real-time identification and compare its F1 score with the other two traditional recognition algorithms. The experiment shows that the F1 score of the DenseNet-40 is up to 0.96, which indicates the DenseNet-40 is provably higher than other traditional algorithms in accuracy. Through trial and error, we find that the echo intensity of lidar can clearly show the texture information of coal and gangue. After combining with the DenseNet-40, it has more benefits than the existing classification methods in accuracy, efficiency, and robustness.

Full Article | PDF Article

More Like This

Target classification of multislit streak tube imaging lidar based on deep learning

Wenhao Li, Shangwei Guo, Yu Zhai, Fei Liu, Zhengchao Lai, and Shaokun Han
Appl. Opt. 60(28) 8809-8817 (2021)

Low-complexity adaptive radius outlier removal filter based on PCA for lidar point cloud denoising

Yao Duan, Chuanchuan Yang, and Hongbin Li
Appl. Opt. 60(20) E1-E7 (2021)

Denoising method for a lidar bathymetry system based on a low-rank recovery of non-local data structures

Bin Hu, Yiqiang Zhao, Rui Chen, Qiang Liu, Pinquan Wang, and Qi Zhang
Appl. Opt. 61(1) 69-76 (2022)

Previous Article Next Article

Supplementary Material (1)

Name	Description
Supplement 1	Supplemental document

Data Availability

Data underlying the results presented in this paper are not publicly available at this time but may be obtained from the authors upon reasonable request.

Cited By

You do not have subscription access to this journal. Cited by links are available to subscribers only. You may subscribe either as an Optica member, or as an authorized user of your institution.

Contact your librarian or system administrator
or
Login to access Optica Member Subscription

Figures (7)

You do not have subscription access to this journal. Figure files are available to subscribers only. You may subscribe either as an Optica member, or as an authorized user of your institution.

Contact your librarian or system administrator
or
Login to access Optica Member Subscription

Tables (3)

You do not have subscription access to this journal. Article tables are available to subscribers only. You may subscribe either as an Optica member, or as an authorized user of your institution.

Contact your librarian or system administrator
or
Login to access Optica Member Subscription

Equations (6)

You do not have subscription access to this journal. Equations are available to subscribers only. You may subscribe either as an Optica member, or as an authorized user of your institution.

Contact your librarian or system administrator
or
Login to access Optica Member Subscription

Materials	Reflectivity (%)
White masonry	85
Limestone, clay	75
Carbonate sand	41
Lava	8

Layer	Output Size	$K = 12$
Convolution	$112 \times 112$	$7 \times 7 c o n v$ , stride2
Pooling	$56 \times 56$	$3 \times 3 max p o o l$ , stride 2
Dense block (1)	$56 \times 56$	$[\begin{matrix} 1 \times 1 c o n v \\ 3 \times 3 c o n v \end{matrix}] \times 6$
Transition layer (1)	$56 \times 56$	$1 \times 1 c o n v$
Transition layer (1)	$28 \times 28$	$2 \times 2 a v e r a g e p o o l$ , stride 2
Dense block (2)	$28 \times 28$	$[\begin{matrix} 1 \times 1 c o n v \\ 3 \times 3 c o n v \end{matrix}] \times 12$
Transition layer (2)	$28 \times 28$	$1 \times 1 c o n v$
Transition layer (2)	$14 \times 14$	$2 \times 2 a v e r a g e p o o l$ , stride 2
Dense block (3)	$14 \times 14$	$[\begin{matrix} 1 \times 1 c o n v \\ 3 \times 3 c o n v \end{matrix}] \times 48$
Transition layer (3)	$14 \times 14$	$1 \times 1 c o n v$
Transition layer (3)	$7 \times 7$	$2 \times 2 a v e r a g e p o o l$ , stride 2
Dense block (4)	$7 \times 7$	$[\begin{matrix} 1 \times 1 c o n v \\ 3 \times 3 c o n v \end{matrix}] \times 32$
Classification layer	$1 \times 1$	$7 \times 7$ global average pool
		1000D fully connected, softmax

Model	Accuracy (%)	Precision	Recall	F1 Score	Speed (EIIs/s)
GLCM	70.10	0.69	0.71	0.70	29.28
HOG-LBP	85.18	0.86	0.85	0.85	26.32
DenseNet-121	96.68	0.96	0.97	0.96	41.69
DenseNet-40 (ours)	95.93	0.95	0.97	0.96	67.05

Materials	Reflectivity (%)
White masonry	85
Limestone, clay	75
Carbonate sand	41
Lava	8

Layer	Output Size	$K = 12$
Convolution	$112 \times 112$	$7 \times 7 c o n v$ , stride2
Pooling	$56 \times 56$	$3 \times 3 max p o o l$ , stride 2
Dense block (1)	$56 \times 56$	$[\begin{matrix} 1 \times 1 c o n v \\ 3 \times 3 c o n v \end{matrix}] \times 6$
Transition layer (1)	$56 \times 56$	$1 \times 1 c o n v$
Transition layer (1)	$28 \times 28$	$2 \times 2 a v e r a g e p o o l$ , stride 2
Dense block (2)	$28 \times 28$	$[\begin{matrix} 1 \times 1 c o n v \\ 3 \times 3 c o n v \end{matrix}] \times 12$
Transition layer (2)	$28 \times 28$	$1 \times 1 c o n v$
Transition layer (2)	$14 \times 14$	$2 \times 2 a v e r a g e p o o l$ , stride 2
Dense block (3)	$14 \times 14$	$[\begin{matrix} 1 \times 1 c o n v \\ 3 \times 3 c o n v \end{matrix}] \times 48$
Transition layer (3)	$14 \times 14$	$1 \times 1 c o n v$
Transition layer (3)	$7 \times 7$	$2 \times 2 a v e r a g e p o o l$ , stride 2
Dense block (4)	$7 \times 7$	$[\begin{matrix} 1 \times 1 c o n v \\ 3 \times 3 c o n v \end{matrix}] \times 32$
Classification layer	$1 \times 1$	$7 \times 7$ global average pool
		1000D fully connected, softmax

Abstract

Supplementary Material (1)

Data Availability

Cited By

Figures (7)

Tables (3)

Equations (6)

Applied Optics