Object Detection Algorithm in Remote Sensing Images Based on Improved YOLOX

Zhaohua Hu; Yuhui Li

doi:10.3788/LOP231615

Laser & Optoelectronics Progress, Volume. 61, Issue 12, 1228004(2024)

Object Detection Algorithm in Remote Sensing Images Based on Improved YOLOX

Zhaohua Hu^1,2 and Yuhui Li^1、*

¹School of Electronic and Information Engineering, Nanjing University of Information Science and Technology, Nanjing 210044, Jiangsu , China

²Jiangsu Collaborative Innovation Center on Atmospheric Environment and Equipment Technology, Nanjing University of Information Science and Technology, Nanjing 210044, Jiangsu , China

show less

Abstract Get PDF(in Chinese)

Figures & Tables(16)

Fig. 1. Structure of the YOLOX algorithm

Download full size

Fig. 2. Structure of improved YOLOX algorithm

Download full size

Fig. 3. Structure of RCAM

Download full size

Fig. 4. Structure of DM

Download full size

Fig. 5. Structure of FFM. (a) Branch 1; (b) branch 2; (c) branch 3

Download full size

Fig. 6. Structure of attention mechanism CAS

Download full size

Fig. 7. Structure of FEM

Download full size

Fig. 8. Partial images and targets on DIOR and RSOD datasets

Download full size

Fig. 9. Renderings before and after preprocessing. (a) Original images; (b) images after ACE preprocessing

Download full size

Fig. 10. mAP50 radar chart of algorithms before and after improvement under different categories

Download full size

Fig. 11. Visualization results of thermal diagrams. (a) YOLOX; (b) improved YOLOX

Download full size

Fig. 12. Comparison charts of detection results. (a) Ground truth in remote sensing images; (b) YOLOX; (c) improved YOLOX

Download full size

Table 1. Ablation experimental results of the proposed algorithm on the DIOR dataset

View table

Table 1. Ablation experimental results of the proposed algorithm on the DIOR dataset

Method	Params /10⁶	GFLOPs /10⁹	AP_S /%	AP_M /%	AP_L /%	AP /%	FPS /（frame/s）	mAP50 /%
YOLOX	5.04	15.2	11.3	35.3	62.7	43.9	56	69.79
YOLOX+ACE	5.04	15.2	11.5	37.9	66.4	47.0	56	72.71
YOLOX+RCAM	5.10	15.3	12.1	38.6	66.1	47.1	55	72.84
YOLOX+FFM	6.25	17.4	12.3	39.6	67.1	47.5	51	73.22
YOLOX+FEM	5.41	15.7	11.8	37.7	66.7	47.4	53	73.08
YOLOX+RCAM+FFM+FEM	6.69	18.8	12.6	40.1	67.4	48.5	50	73.87

Table 2. Comparative experimental results of algorithms with different attention machnisms

View table

Table 2. Comparative experimental results of algorithms with different attention machnisms

Method	Params /10⁶	AP_S /%	AP_M /%	AP_L /%	AP /%	FPS /（frame/s）	mAP50 /%
YOLOX+CBAM	5.42	11.7	36.8	66.6	47.2	51	72.90
YOLOX+SimAM	5.40	11.5	37.6	66.5	47.3	55	72.86
YOLOX+CA	5.41	11.6	38.0	66.6	47.1	54	73.01
YOLOX+CAS	5.41	11.8	37.7	66.7	47.4	53	73.08

Table 3. Comparative experimental results of different algorithms on the DIOR dataset

View table

Table 3. Comparative experimental results of different algorithms on the DIOR dataset

Method	Size	Params /10⁶	GFLOPs /10⁹	FPS /（frame/s）	mAP50 /%	AP_S /%	AP_M /%	AP_L /%
Faster R-CNN	640	28.50	948.4	6.0	63.10	6.5	32.3	57.6
CenterNet	640	32.70	109.3	19.0	56.05	5.4	25.2	51.4
YOLOv4	640	5.90	16.2	67.0	61.01	6.7	31.3	50.5
YOLOv5	640	7.10	16.5	50.0	66.97	11.1	37.4	62.0
YOLOv7	640	6.10	13.3	66.0	72.83	12.3	38.9	69.1
CF2PN	640	91.60	>31.0	19.7	67.25	11.3	36.0	61.4
YOLOX	640	5.04	15.2	56.0	69.79	11.3	35.3	62.7
Improved YOLOX	640	6.69	18.8	50.0	73.87	12.6	40.1	67.4

Table 4. Comparative experimental results of different algorithms on the RSOD dataset

View table

Table 4. Comparative experimental results of different algorithms on the RSOD dataset

Method	Size	Params /10⁶	FPS /（frame/s）	AP /%				mAP50 /%
Method	Size	Params /10⁶	FPS /（frame/s）	Aircraft	Overpass	Oiltank	Playground	mAP50 /%
Faster R-CNN	640	28.50	6	71.90	100.00	90.90	100.00	90.70
CenterNet	640	32.70	19	70.83	85.37	86.56	100.00	85.69
YOLOv4	640	5.90	67	78.28	84.58	83.94	100.00	86.70
YOLOv5	640	7.10	50	94.10	80.50	95.10	99.10	92.20
MDCF²Det	640	—	—	81.02	100.00	90.77	100.00	92.95
YOLOX	640	5.04	56	95.95	86.93	96.64	100.00	94.88
Improved YOLOX	640	6.69	50	97.20	88.33	99.35	100.00	96.22

Tools

Get Citation

Copy Citation Text

Zhaohua Hu, Yuhui Li. Object Detection Algorithm in Remote Sensing Images Based on Improved YOLOX[J]. Laser & Optoelectronics Progress, 2024, 61(12): 1228004

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: Jun. 27, 2023

Accepted: Aug. 30, 2023

Published Online: May. 20, 2024

The Author Email: Yuhui Li (523700486@qq.com)

DOI:10.3788/LOP231615

CSTR:32186.14.LOP231615

Topics

laser devices and laser physics

Lasers and Laser Optics

Laser physics

laser manufacturing

Instrumentation, Measurement and Metrology