Target Detection Method for Remote Sensing Images Based on Sparse Mask Transformer

Xulun Liu; Shiping Ma; Linyuan He; Chen Wang; Xu He; Zhe Chen

doi:10.3788/LOP202259.2228005

Laser & Optoelectronics Progress, Volume. 59, Issue 22, 2228005(2022)

Target Detection Method for Remote Sensing Images Based on Sparse Mask Transformer

Xulun Liu¹, Shiping Ma¹, Linyuan He^1,2、*, Chen Wang¹, Xu He¹, and Zhe Chen³

Author Affiliations

¹School of Aeronautical Engineering, Air Force Engineering University, Xi'an 710038, Shaanxi, China

²Unbanned System Research Institute, Northwestern Polytechnical University, Xi'an 710072, Shaanxi, China

³School of Cyberspace Security, Xi'an University of Posts & Telecommunications, Xi'an 710121, Shaanxi, China

show less

Abstract Get PDF(in Chinese)

Figures & Tables(8)

Fig. 1. Structure diagram of the proposed network

Download full size

Fig. 2. Schematic of attention block. MSA is multi-head attention, SI-MSA is sparse-interpolation multi-head attention.(a) Standard attention block; (b) sparse-interpolation attention block

Download full size

Fig. 3. Sparse-interlation multi-head self-attention

Download full size

Fig. 4. Deterministic sampling and stochastic sampling. (a) Deterministic sampling; (b) stochastic sampling

Download full size

Fig. 5. Visualization of parts of the detection results

Download full size

Table 1. Comparison of detection accuracy of different methods in DOTA dataset

View table

Table 1. Comparison of detection accuracy of different methods in DOTA dataset

Method	AP /%															mAP /%
Method	PL	BD	BR	GTF	SV	LV	SH	TC	BC	ST	SBF	RA	HA	SP	HC	mAP /%
R2CNN^［19］	80.94	65.67	35.34	67.44	59.92	50.91	55.81	90.67	66.92	72.39	55.06	52.23	55.14	53.35	48.22	60.67
RRPN^［20］	88.52	71.20	31.66	59.30	51.85	56.19	57.25	90.81	72.84	67.38	56.69	52.84	53.08	51.94	53.58	61.01
RT^［21］	88.64	78.52	43.44	75.92	68.81	73.68	83.59	90.74	77.27	81.46	58.39	53.54	62.83	58.93	47.64	69.56
CAD-Net^［2］	87.80	82.40	49.40	73.50	71.10	64.50	76.60	90.90	79.20	73.30	48.40	60.90	62.00	67.00	62.20	69.90
SCRDet^［22］	89.98	80.65	52.09	68.36	68.36	60.32	72.41	90.85	87.94	86.86	65.02	66.68	66.25	68.24	65.21	72.61
GV^［3］	89.64	85.00	52.26	77.34	73.01	73.14	86.82	90.74	79.02	86.81	59.55	70.91	72.94	70.86	57.32	75.02
BBAVectors^［23］	88.63	84.06	52.13	69.56	78.26	80.40	88.06	90.87	87.23	86.39	56.11	65.52	67.10	72.08	63.96	75.36
Proposed method	89.14	84.40	54.73	76.80	79.21	82.01	89.23	91.34	86.05	88.54	68.65	69.90	70.83	74.27	71.37	78.43

Table 2. mAP value and detection speed of different detection methods on DOTA dataset

View table

Table 2. mAP value and detection speed of different detection methods on DOTA dataset

Model	Backbone	mAP /%	Speed /（frame·s^-1）
R2CNN	VGG-16	60.67	5.9
RRPN	VGG-16	61.01	7.2
RT	R101-FPN	69.56	7.8
CAD-Net	R101-FPN	69.90	7.9
SCRDet	R101-FPN	72.61	8.4
GV	R101-FPN	75.02	11.6
BBAVectors	ResNet-101	75.36	13.7
Proposed method	ResNet-101	78.43	12.5

Table 3. Ablation study
View table
Table 3. Ablation study
Baseline Multi-scale input Sampling module Interpolation module Epoch GFLOPs mAP /%
√ 50 152 65.33
√ 500 152 76.41
√ √ 50 1890 67.18
√ √ 500 1890 78.23
√ √ √ 50 138 77.86
√ √ √ √ 50 140 78.43

Tools

Get Citation

Copy Citation Text

Xulun Liu, Shiping Ma, Linyuan He, Chen Wang, Xu He, Zhe Chen. Target Detection Method for Remote Sensing Images Based on Sparse Mask Transformer[J]. Laser & Optoelectronics Progress, 2022, 59(22): 2228005

Download Citation

EndNote(RIS)BibTex Plain Text

Set citation alerts for article

Save article for my favorites