Remote sensing image target detection based on YOLOv5 network

Hao Wu; Li Shui; Chong Zhang; Bingfa Miao

doi:10.1117/12.2617304

22 July 2022 Remote sensing image target detection based on YOLOv5 network

Hao Wu, Li Shui, Chong Zhang, Bingfa Miao

Proceedings Volume 12277, 2021 International Conference on Optical Instruments and Technology: Optical Systems, Optoelectronic Instruments, Novel Display, and Imaging Technology; 1227711 (2022) https://doi.org/10.1117/12.2617304
Event: 2021 International Conference on Optical Instruments and Technology, 2022, Online Only

Abstract

An improved target recognition algorithm based on YOLOv5 network was proposed to solve the problem of low accuracy of target recognition due to the complex background of remote sensing image, small difference between target classes and multiple and dense targets which in the target detection task of optical remote sensing image taken from the overhead angle.The algorithm improves the target recognition by changing the visual activation function, loss function and improving the structure of feature pyramid network (FPN). Firstly, the original YOLOv5 network structure is analyzed that the key technologies of input, Backbone, neck and output are introduced. The LeakyReLU activation function has been changed to FReLU which has the ability to adaptively obtain the local context of the image and is simple in form that improving the spatial sensitivity in the activation function. PIOU loss function is used to replace common CIOU and GIOU that rotation parameters are added to better detect rotating and dense objects. Feature Pyramid Grids (FPG) are introduced that the Feature scale space is represented as a regular grid with parallel bottom-up paths which fused by multi-directional horizontal connections. The single path feature pyramid network is improved by significantly improving its performance at similar computational cost. Experimental results show that under the same training conditions, compared with the original YOLOv5 network, the improved YOLOv5 network converges the training results faster, and the average recognition rate of the training model increases by 5%. Through the test set verification that the recognition accuracy of all kinds of images has been improved which the average accuracy has reached 0.702, 7% higher than the original YOLOv5 network.

Citation Download Citation

Hao Wu, Li Shui, Chong Zhang, and Bingfa Miao "Remote sensing image target detection based on YOLOv5 network", Proc. SPIE 12277, 2021 International Conference on Optical Instruments and Technology: Optical Systems, Optoelectronic Instruments, Novel Display, and Imaging Technology, 1227711 (22 July 2022); https://doi.org/10.1117/12.2617304

ACCESS THE FULL ARTICLE

INSTITUTIONAL
Select your institution to access the SPIE Digital Library.

SELECT YOUR INSTITUTION

PERSONAL
Sign in with your SPIE account to access your personal subscriptions or to use specific features such as save to my library, sign up for alerts, save searches, etc.

PERSONAL SIGN IN

No SPIE Account? Create one

PURCHASE THIS CONTENT

SUBSCRIBE TO DIGITAL LIBRARY

50 downloads per 1-year subscription

Members: $195

Non-members: $335 ADD TO CART

25 downloads per 1 - year subscription

Members: $145

Non-members: $250 ADD TO CART

PURCHASE SINGLE ARTICLE

Includes PDF, HTML & Video, when available

Members: $17.00

Non-members: $21.00 ADD TO CART

PROCEEDINGS
8 PAGES

DOWNLOAD PAPER SAVE TO MY LIBRARY

GET CITATION

RIGHTS & PERMISSIONS

Get copyright permission Get copyright permission on Copyright Marketplace

KEYWORDS

Target detection

Target recognition

Detection and tracking algorithms

Remote sensing

Neck

Convolution

Image enhancement

Show All Keywords

Keywords/Phrases

Search In:

Publication Years