Multiscale Object Detection Algorithm for Satellite Remote-Sensing Images

Jianhong Xiang; Zhenxing Chen; Linyu Wang

doi:10.3788/LOP212670

Journals >Laser & Optoelectronics Progress >Volume 60 >Issue 2 >Page 0228003 > Article

Laser & Optoelectronics Progress
Vol. 60, Issue 2, 0228003 (2023)

Multiscale Object Detection Algorithm for Satellite Remote-Sensing Images

Jianhong Xiang^1,2, Zhenxing Chen^1,2, and Linyu Wang^1,2,*

Author Affiliations

¹College of Information & Communication Engineering, Harbin Engineering University, Harbin 150001, Heilongjiang, China

²Key Laboratory of Advanced Ship Communication and Information Technology, Harbin Engineering University, Harbin 150001, Heilongjiang, China

show less

DOI: 10.3788/LOP212670 Cite this Article Set citation alerts

Jianhong Xiang, Zhenxing Chen, Linyu Wang. Multiscale Object Detection Algorithm for Satellite Remote-Sensing Images[J]. Laser & Optoelectronics Progress, 2023, 60(2): 0228003 Copy Citation Text

show less

Abstract

A multiscale object detection algorithm for satellite remote-sensing images is proposed to solve the problems of background confusion, low precision of small object detection, and high miss rate in multiscale object detection. The channel and spatial attention module is used in the backbone network, and the feature fusion network is redesigned to realize the multiple fusion of up-down-up sampling. The channel weight parameter is added to enable the network to pay more attention to critical channels, fully utilize different feature information levels, and enhance the detailed feature information. In a DIOR dataset, not only the detection effect of small objects but also the detection accuracy of objects in complex scenes is improved. Compared with that using YOLOv5m, the detection effect of some small or complex objects is improved significantly, the accuracy is improved by more than 4.5 percentage points, and the overall accuracy is improved by 3.1 percentage points.

Keywords

attention mechanism channel weight feature fusion multiscale object detection neural network remote sensing

\{\begin{matrix} L_{C I o U} = 1 - R_{I o U} + \frac{ρ^{2} (B, G)}{c^{2}} + α v \\ R_{I o U} = \frac{|G ⋂ B|}{|G ⋃ B|} \end{matrix},

（1）

\{\begin{matrix} v = \frac{4}{π^{2}} {(a r c t a n \frac{w_{g t}}{h_{g t}} - a r c t a n \frac{w}{h})}^{2} \\ α = \frac{v}{(1 - R_{I o U}) + v} \end{matrix},

（2）

\{\begin{matrix} L = #x1D4C1 (x, y) = {l_{1}, \dots, l_{n}, \dots l_{N}}^{T} \\ l_{n} = - w_{n} \{y_{n} \cdot l o g σ (x_{n}) + (1 - y_{n}) \cdot l o g [1 - σ (x_{n})]\} \end{matrix},

（3）

y_{C} = σ \{f_{M L P} [A v g P o o l (x)] + f_{M L P} [M a x P o o l (x)]\},

（4）

y_{S} = σ \{f^{7 \times 7} [A v g P o o l (x); M a x P o o l (x)]\}

，（5）

\begin{matrix} P_{4 o u t} = \\ C o n v [\frac{w_{1} \cdot u p s a m p l e (P_{5 o u t}) + w_{2} \cdot P_{4 l a s t} + w_{3} \cdot P_{4 b a c k b o n e}}{w_{1} + w_{2} + w_{3} + ε}] 。 \end{matrix}

（6）

\{\begin{matrix} P = \frac{N_{T P}}{N_{T P} + N_{F P}} \\ R = \frac{N_{T P}}{N_{T P} + N_{F N}} \\ P_{A P} = \int_{0}^{1} P (R) d R \\ P_{m A P} = \frac{\sum_{i = 1}^{N} P_{A P i}}{N} \end{matrix},

（7）

Jianhong Xiang, Zhenxing Chen, Linyu Wang. Multiscale Object Detection Algorithm for Satellite Remote-Sensing Images[J]. Laser & Optoelectronics Progress, 2023, 60(2): 0228003

Download Citation

Set citation alerts for the article

Tools

Set citation alerts for the article

Save the article for my favorites

Paper Information

微信扫一扫：分享

微信扫一扫：分享