Small Object Detection in Remote Sensing Images Based on Feature Fusion and Attention

Yin Zhang; Guiyi Zhu; Tianjun Shi; Kun Zhang; Junhua Yan

doi:10.3788/AOS202242.2415001

Acta Optica Sinica, Volume. 42, Issue 24, 2415001(2022)

Small Object Detection in Remote Sensing Images Based on Feature Fusion and Attention

Yin Zhang^1,2, Guiyi Zhu^1,2, Tianjun Shi³, Kun Zhang^1,2, and Junhua Yan^1,2、*

¹Space Photoelectric Detection and Sensing of Industry and Information Technology, Nanjing University of Aeronautics and Astronautics, Nanjing 211106, Jiangsu , China

²College of Astronautics, Nanjing University of Aeronautics and Astronautics, Nanjing 211106, Jiangsu , China

³Research Center for Space Optical Engineering, Harbin Institute of Technology, Harbin 150001, Heilongjiang , China

show less

Abstract Get PDF(in Chinese)

Figures & Tables(14)

Fig. 1. Overall scheme of FFAM-YOLO algorithm

Download full size

Fig. 2. Feature enhancement module FEM

Download full size

Fig. 3. Feature fusion module FFM

Download full size

Fig. 4. Attention module ESM

Download full size

Fig. 5. Examples of image offical annotation

Download full size

Fig. 6. Examples of manual supplementary annotation

Download full size

Fig. 7. Distribution of small object ground truth box size

Download full size

Fig. 8. Distribution of numbers of small objects

Download full size

Fig. 9. Detection results of YOLOv5 and FFAM-YOLO in driving scene and parking scene

Download full size

Fig. 10. Detection results of YOLOv5 and FFAM-YOLO for low light scene and occlusion scene

Download full size

Fig. 11. FFAM-YOLO detection misjudgment situation

Download full size

Fig. 12. Detection results of contrast experiments using FFAM-YOLO

Download full size

Table 1. Evaluation results of each module in proposed algorithm
View table
Table 1. Evaluation results of each module in proposed algorithm
FEM&FFM ESM Algorithm P /% R /% AP for IoU of 0.5 /%
AP for IoU of
0.5∶0.95 /%
× × YOLOv5 90.3 81.2 87.1 31.6
√ × FFAM-YOLO 91.4 83.4 88.2 32.0
√ √ FFAM-YOLO 91.9 83.5 89.0 32.6

Table 2. Comparison of detetion results of FFAM-YOLO and other five algorithms

View table

Table 2. Comparison of detetion results of FFAM-YOLO and other five algorithms

Algorithm	Backbone	P /%	R /%	AP for IoU of 0.5 /%	AP for IoU of 0.5∶0.95 /%	Detection rate /（frame·s^-1）
DSSD	ResNet101	64.6	57.4	53.1	16.3	34.3
YOLOv3	DarkNet53	72.0	69.1	57.4	18.5	61.1
YOLOv4	DarkNet53	81.0	82.0	77.8	28.8	61.2
RefineDet	ResNet101	88.1	82.3	79.1	31.4	31.3
YOLOv5	CSPDarkNet53	90.3	81.2	87.1	31.6	181.9
FFAM-YOLO	CSPDarkNet53	91.9	83.5	89.0	32.6	120.0

Tools

Get Citation

Copy Citation Text

Yin Zhang, Guiyi Zhu, Tianjun Shi, Kun Zhang, Junhua Yan. Small Object Detection in Remote Sensing Images Based on Feature Fusion and Attention[J]. Acta Optica Sinica, 2022, 42(24): 2415001

Download Citation

EndNote(RIS)BibTex Plain Text

Set citation alerts for article

Save article for my favorites