Multiscale Convolutional Neural Network-Based Lithology Classification Method for Multisource Data Fusion

Song Dai; Ximing Sun; Jingming Zhang; Yongshan Zhu; Bin Wang; Dongmei Song

doi:10.3788/LOP232491

Laser & Optoelectronics Progress, Volume. 61, Issue 14, 1437005(2024)

Multiscale Convolutional Neural Network-Based Lithology Classification Method for Multisource Data Fusion

Song Dai^1、*, Ximing Sun², Jingming Zhang², Yongshan Zhu², Bin Wang¹, and Dongmei Song¹

Author Affiliations

¹College of Ocean and Space Information, China University of Petroleum (East China), Qingdao 266580, Shandong , China

²Bureau of Geophysical Prospecting INC., China National Petroleum Corporation, Zhuozhou072750, Hebei , China

show less

Abstract Get PDF(in Chinese)

Figures & Tables(15)

Fig. 1. Lithologic classification network flow chart of multisource data fusion based on multiscale CNN

Download full size

Fig. 2. Multiscale CNN feature extraction module

Download full size

Fig. 3. Convolutional attention feature fusion module

Download full size

Fig. 4. Schematic diagram of MSCA attention module

Download full size

Fig. 5. CNN module structure diagram

Download full size

Fig. 6. Main geological types in the study area. (a) N₁t; (b) N₁s; (c) E_2-3a; (d) N₂d; (e) Q₁x; (f) Q

Download full size

Fig. 7. Multisource data preprocessing flow chart

Download full size

Fig. 8. Visualization of Sichuan basin data. (a) Colour image after sharpening of aerial remote sensing images; (b) grayscale image of DEM data; (c) reference map of geological boundaries; (d) lithology label

Download full size

Fig. 9. Results of lithology classification in the study area. (a) Color image of aerial remote sensing image; (b) lithology label; (c) KNN classification map; (d) MLC classification map; (e) SVM classification map; (f) RF classification map; (g) aerial remote sensing image classification map; (h) fusion classification map

Download full size

Table 1. Sichuan basin data parameter description table
View table
Table 1. Sichuan basin data parameter description table
Location Sichuan， China
Aircraft type Cessna-208
Sensor name Phase IXU-RS RIEGL LMS-Q1560
Sensor type RGB DEM
Image size /（pixel×pixel） 5074×3678
Spatial resolution /m 5
Number of bands 3 1

Table 2. Comparison table of results of ablation experiment evaluation
View table
Table 2. Comparison table of results of ablation experiment evaluation
Dataset Multiscale CNN MSCA $O_{O A}$ /% $A_{A A}$ /% $K_{K a p p a}$
Sichuan basin data √ 88.26 88.44 0.8460
√ 89.79 86.87 0.8658
√ √ 92.22 90.35 0.8966

Table 3. Comparison of classification performance of different dilation rates
View table
Table 3. Comparison of classification performance of different dilation rates
Method Dilation rate $O_{O A}$ /% $A_{A A}$ /% $K_{K a p p a}$
Singlescale ｛3，3，3｝ 89.82 88.31 0.8532
Multiscale large size ｛2，4，6｝ 90.56 88.72 0.8763
Propose method ｛1，3，5｝ 92.22 90.35 0.8966

Table 4. Comparison of classification performance of different fusion methods
View table
Table 4. Comparison of classification performance of different fusion methods
Method Feature size $O_{O A}$ /% $A_{A A}$ /% $K_{K a p p a}$
Concatenation 1×1×512 91.24 89.52 0.8829
Element-wise add 1×1×256 92.22 90.35 0.8966

Table 5. Comparison table of model complexity and execution time for different modules
View table
Table 5. Comparison table of model complexity and execution time for different modules
Dataset Multiscale CNN MSCA Params /10⁶ FLOPs /10⁶ Execution time /s
Sichuan basin data √ 0.93 21.20 1291.16
√ 0.69 5.72 410.71
√ √ 0.94 21.23 1322.70

Table 6. Accuracy of lithology classification by different methods in Sichuan basin

View table

Table 6. Accuracy of lithology classification by different methods in Sichuan basin

No.	Class	KNN	MLC	SVM	RF	Propose（RGB）	Propose（RGB+DEM）
1	N₁t	64.08	76.79	99.56	99.71	99.96	98.37
2	N₁s	94.93	90.45	96.47	98.21	91.88	97.88
3	E_2-3a	87.09	99.43	84.68	83.60	99.72	99.87
4	N₂d	65.87	85.02	80.66	99.06	67.40	87.25
5	Q₁x	68.53	51.00	60.90	62.28	61.84	69.55
6	Q	56.28	82.00	82.62	65.54	88.11	89.19
$O_{O A}$ /%		68.60	78.86	80.40	82.53	87.97	92.22
$A_{A A}$ /%		72.80	80.78	84.15	84.73	84.82	90.35
$K_{K a p p a}$		0.5316	0.6617	0.7537	0.7789	0.8402	0.8966

Tools

Get Citation

Copy Citation Text

Song Dai, Ximing Sun, Jingming Zhang, Yongshan Zhu, Bin Wang, Dongmei Song. Multiscale Convolutional Neural Network-Based Lithology Classification Method for Multisource Data Fusion[J]. Laser & Optoelectronics Progress, 2024, 61(14): 1437005

Download Citation

EndNote(RIS)BibTex Plain Text

Set citation alerts for article

Save article for my favorites