Lithology Segmentation Method with Efficient Channel Attention Based on Multiple Eigenvalues of Outcrop Voxel

Lang Liu; Yanlin Shao; Qihong Zeng; Kunpeng Zhao; Changhui Zhou; Peijin Li; Rui Zeng

doi:10.3788/LOP241999

Laser & Optoelectronics Progress, Volume. 62, Issue 10, 1037002(2025)

Lithology Segmentation Method with Efficient Channel Attention Based on Multiple Eigenvalues of Outcrop Voxel

Lang Liu¹, Yanlin Shao^1、*, Qihong Zeng², Kunpeng Zhao¹, Changhui Zhou¹, Peijin Li¹, and Rui Zeng¹

Author Affiliations

¹School of Geosciences, Yangtze University, Wuhan 430100, Hubei , China

²Research Institute of Petroleum Exploration & Development, Beijing 100000, China

show less

Abstract Get PDF(in Chinese)

Figures & Tables(11)

Fig. 1. Structure of MGECA network

Download full size

Fig. 2. Structure of multi-granularity convolutional neural network

Download full size

Fig. 3. Structure of dual-channel group convolution module

Download full size

Fig. 4. Structure of ECA-Transformer encoder

Download full size

Fig. 5. Camera photograph and calibrated laser scan result of outcrop in Crescent Bay

Download full size

Fig. 6. Schematic diagram of optimal parameter separability

Download full size

Fig. 7. Comparison results of ablation experiments

Download full size

Fig. 8. Comparison of identification results for laser outcrop dataset in Crescent Bay

Download full size

Table 1. Ablation experimental results of MGECA
View table
Table 1. Ablation experimental results of MGECA
Baseline Multi-granularity convolution Dual-channel group convolution ECA OA mAcc mIoU
√ 84.1 80.2 58.8
√ √ 86.2 81.6 63.5
√ √ √ 88.8 83.8 68.2
√ √ √ √ 90.6 85.7 70.4

Table 2. Results of comparative experiments on laser outcrop datasets in Crescent Bay
View table
Table 2. Results of comparative experiments on laser outcrop datasets in Crescent Bay
Method OA mAcc mIoU
SVM 56.7 50.5 37.5
DGPoint 58.9 53.4 45.7
DGCNN 85.1 72.4 56.1
DBDA 88.0 78.4 63.5
MGECA 90.6 85.7 70.4

Table 3. Identification accuracy of single target type by different methods

View table

Table 3. Identification accuracy of single target type by different methods

Method	Lithology	AA /%	Every epoch training time /s
SVM	Sandstone	26.7	75
	Conglomerate	40.1
	Mudstone	57.2
	Grass	62.1
DGPoint	Sandstone	35.9	164
	Conglomerate	79.8
	Mudstone	68.4
	Grass	73.1
DGCNN	Sandstone	11.2	235
	Conglomerate	94.1
	Mudstone	65.6
	Grass	78.4
DBDA	Sandstone	67.2	110
	Conglomerate	84.6
	Mudstone	70.9
	Grass	89.4
MGECA	Sandstone	74.3	131
	Conglomerate	91.6
	Mudstone	89.2
	Grass	58.6

Tools

Get Citation

Copy Citation Text

Lang Liu, Yanlin Shao, Qihong Zeng, Kunpeng Zhao, Changhui Zhou, Peijin Li, Rui Zeng. Lithology Segmentation Method with Efficient Channel Attention Based on Multiple Eigenvalues of Outcrop Voxel[J]. Laser & Optoelectronics Progress, 2025, 62(10): 1037002

Download Citation

EndNote(RIS)BibTex Plain Text

Set citation alerts for article

Save article for my favorites