Small Ship Target Detection Method for Remote Sensing Images Based on Dual Feature Enhancement

Fig. 11. Result comparison between proposed method and R³Det benchmark experiment.(a1)(a2) Texture enhancement; (b1)(b2) color enhancement; (c1)(c2) angle alignment

Download full size

Table 1. Action color adjustment operation

View table

Table 1. Action color adjustment operation

Action code	Action description	Action code	Action description
1	Contrast（0.95×）	2	Contrast（1.05×）
3	Color saturation（0.95×）	4	Color saturation（1.05×）
5	Brightness（0.95×）	6	Brightness（1.05×）
7	Red+green（0.95×）	8	Red+green（1.05×）
9	Green+blue（0.95×）	10	Green+blue（1.05×）
11	Red+blue（0.95×）	12	Red+blue（1.05×）

Table 2. Comparative experimental results in DOTA data set
View table
Table 2. Comparative experimental results in DOTA data set
Method mAP /%
RetinaNet^［28］ 65.73
CSL^［20］ 67.38
RsDet^［24］ 66.05
RiDet^［25］ 66.06
SCRDet^［27］ 71.73
R³Det^［26］ 70.66
Proposed（DEGAN） 72.87

Table 3. Comparative experimental results in HRSC2016 data set

View table

Table 3. Comparative experimental results in HRSC2016 data set

Method	mAP /%	Speed /（frame·s^-1）
R²CNN^［30］	72.98	6.00
RetinaNet^［28］	82.57	19.48
RiDet^［25］	83.78	25.23
RsDet^［24］	83.47	25.72
R³Det^［26］	86.52	20.27
ReDet^［30］	88.06	4.20
Proposed（DEGAN）	89.91	16.37

Table 4. Results of ablation experiment
View table
Table 4. Results of ablation experiment
Network Module 1 Module 2 Module 3 Module 4
mAP /% 87.45 87.92 89.04 89.91
TEN √ √ √ √
CEN √ √ √
ATFPN √ √
CSL √

Tools

Get Citation

Copy Citation Text

Zhijing Xu, Xue Bai. Small Ship Target Detection Method for Remote Sensing Images Based on Dual Feature Enhancement[J]. Acta Optica Sinica, 2022, 42(18): 1828002

Download Citation

EndNote(RIS)BibTex Plain Text

Set citation alerts for article

Save article for my favorites