Land Cover Classification of UAV Visible Remote Sensing Based on Joint Distribution of Color-Spatial Feature

Yushuang Zeng; Shaohua Zeng; Li Yuan; Ying Long

doi:10.3788/LOP240511

Laser & Optoelectronics Progress, Volume. 61, Issue 24, 2428004(2024)

Land Cover Classification of UAV Visible Remote Sensing Based on Joint Distribution of Color-Spatial Feature

Yushuang Zeng^1,2、*, Shaohua Zeng^1,2, Li Yuan³, and Ying Long⁴

Author Affiliations

¹College of Computer and Information Science, Chongqing Normal University, Chongqing 401331, China

²Chongqing Research Center on Engineer Technology of Digital Agricultural & Services, Chongqing 401331, China

³College of Information Engineering, Chongqing Electric Power College, Chongqing 400053, China

⁴College of Intelligent Information Engineering, Chongqing Aerospace Polytechnic College, Chongqing 400022, China

show less

Abstract Get PDF(in Chinese)

Figures & Tables(14)

Fig. 1. Framework of land cover classification

Download full size

Fig. 2. Logarithmic spiral by a golden rectangle

Download full size

Fig. 3. Demonstration of the PDF of $Δ E'$ , $Δ W'$ , $Δ W_{1}$ , $Δ E_{1}$ , $Δ W_{2}$ , and $Δ E_{2}$ , respectively

Download full size

Fig. 4. Correlation coefficient of $(Δ E_{1}, Δ W_{1}, Δ E_{2}, Δ W_{2}, Δ E', Δ W')$ . (a) Rice; (b) corn; (c) forest

Download full size

Fig. 5. Process for solving multidimensional mixed Weibulls distribution

Download full size

Fig. 6. Schematic of research area. (a) Research area 2; (b) orthophoto map; (c) research area 1

Download full size

Fig. 7. Schematic diagram of the selected training sample. (a) (c) Sampling diagram of part of the patch by logarithmic spiral; (b) (d) random sampling diagram of part of the patch (red box identifies adjacent training samples)

Download full size

Fig. 8. Results of the post-processing flow. (a) Original classification result; (b) edge graph of Fig. (a) detected by canny operator; (c) majority filtering result obtained by region growth under constraint of image edge; (d) edge graph of Fig. (c) detected by canny operator; (e) 3×3 majority filtering result under constraint of image edge

Download full size

Fig. 9. Comparison of classification results of different algorithms

Download full size

Fig. 10. Generalization test results of classification algorithms

Download full size

Table 1. Classification results of feature extraction window size and neighborhood size

View table

Table 1. Classification results of feature extraction window size and neighborhood size

Class	Pixel number		Classification result /%
	Pixel number		N=5			N=7			N=9			N=11
	Train	Test	r=1， K=8	r=2， K=16	r=3， K=24	r=1， K=8	r=2， K=16	r=3， K=24	r=1， K=8	r=2， K=16	r=3， K=24	r=1， K=8	r=2， K=16	r=3， K=24
Rice	44	716	100	100	100	99.9	98.7	97.3	95.5	92.6	95.5	100	95.9	87.9
Corn	89	9514	97.1	97.4	97.5	97.7	98.0	97.6	96.1	96.3	95.6	96.9	97.2	96.2
Forest	46	6027	99.9	99.9	99.9	100	99.1	99.4	100	100	100	99.8	99.0	100
OA	‒	‒	98.3	98.5	98.6	98.6	98.4	98.2	97.6	97.5	97.3	98.1	97.8	97.2
AA	‒	‒	99	99.1	99.1	99.2	98.6	98.1	97.2	96.3	97.0	98.9	97.4	94.7
KP	‒	‒	96.7	97.1	97.3	97.4	97.0	96.6	95.5	95.3	95.0	96.5	95.8	94.9

Table 2. Experimental results on effectiveness of training sample selection

View table

Table 2. Experimental results on effectiveness of training sample selection

Pixel number			Proposed /%	Classification result /%
Class	Train	Test	Proposed /%	1	2	3	4	5	6	7	8	9	10	Average
Rice	44	716	99.9	99.1	100	99.9	99.9	99.5	99.9	98.9	99.9	96.9	100	99.4
Corn	89	9514	97.7	98.3	96.6	93.7	93.9	95.6	97.9	97.5	98.4	97.8	98.9	96.9
Forest	46	6027	100	99.9	99.9	100	97.6	99.7	99.2	96.4	98.6	100	99.5	99.0
OA	‒	‒	98.6	99.0	98.2	96.6	95.6	95.3	98.5	97.1	98.4	98.9	99.8	97.7
AA	‒	‒	99.2	99.1	98.8	97.9	97.1	98.3	99.0	97.6	99.0	98.2	99.5	98.5
KP	‒	‒	97.4	98.1	96.6	94.0	91.7	95.1	97.1	94.3	96.9	97.9	99.6	96.1

Table 3. Experimental results of feature extraction and classification algorithms

View table

Table 3. Experimental results of feature extraction and classification algorithms

Class	Pixel number		Classification result /%
Class	Train	Test	Proposed	LBP	GLCM	Basic RF	Improved RF	ResNet	VGG
Rice	44	716	99.9	100	100	99.6	87.8	93.9	66.9
Corn	89	9514	97.7	75.1	53.5	78.0	94.8	99.4	98.2
Forest	46	6027	100	62.5	87.8	71.7	95.4	89.4	84.9
OA	‒	‒	98.6	71.5	68.3	76.6	94.7	95.4	91.9
AA	‒	‒	99.2	79.2	80.4	83.1	92.7	94.2	83.3
KP	‒	‒	97.4	54.1	53.4	59.7	90.0	91.2	83.8

Table 4. Classification results of generalization test of classification algorithms

View table

Table 4. Classification results of generalization test of classification algorithms

Class	Pixel number		Classification result /%
Class	Train	Test	Proposed	LBP	GLCM	Basic RF	Improved RF	ResNet	VGG
Rice	44	503	99.6	98.3	99.1	91.8	96.6	95.5	63.3
Corn	89	873	93.1	86.0	79.8	67.4	89.1	96.9	72.6
Forest	46	1687	100	57.4	100	76.1	97.6	93.9	100
OA	‒	‒	98.0	72.6	94.0	76.2	95.0	95.1	85.9
AA	‒	‒	97.6	80.6	92.9	78.4	94.4	95.4	78.6
KP	‒	‒	96.6	58.3	89.9	61.9	91.5	91.8	75.2

Tools

Get Citation

Copy Citation Text

Yushuang Zeng, Shaohua Zeng, Li Yuan, Ying Long. Land Cover Classification of UAV Visible Remote Sensing Based on Joint Distribution of Color-Spatial Feature[J]. Laser & Optoelectronics Progress, 2024, 61(24): 2428004

Download Citation

EndNote(RIS)BibTex Plain Text

Set citation alerts for article

Save article for my favorites

Paper Information

Category: Remote Sensing and Sensors

Received: Jan. 12, 2024

Accepted: Apr. 26, 2024

Published Online: Dec. 17, 2024

The Author Email:

DOI:10.3788/LOP240511

CSTR:32186.14.LOP240511

Topics

laser devices and laser physics

Lasers and Laser Optics

Laser physics

laser manufacturing

Instrumentation, Measurement and Metrology