Study of Algorithm for Aerial Target Detection Based on Lightweight Neural Network

Real-time detection of targets by video satellites is widely applied for civil and military purposes, but spaceborne platforms are generally limited in memory and computing capacity, with tougher demands on detection algorithms, which can hardly be met by traditional target detection algorithms. The...

Full description

Saved in:
Bibliographic Details
Published in2021 IEEE International Conference on Consumer Electronics and Computer Engineering (ICCECE) pp. 422 - 426
Main Authors Yang, Yumin, Liao, Yurong, Ni, Shuyan, Lin, Cunbao
Format Conference Proceeding
LanguageEnglish
Published IEEE 15.01.2021
Subjects
convolutional neural network
Detection algorithms
lightweight
Neural networks
Object detection
Orbits
parameter
Real-time systems
Satellites
Streaming media
target detection
Online AccessGet full text

Cover

More Information
Summary:Real-time detection of targets by video satellites is widely applied for civil and military purposes, but spaceborne platforms are generally limited in memory and computing capacity, with tougher demands on detection algorithms, which can hardly be met by traditional target detection algorithms. Therefore, this paper proposed a lightweight target detection algorithm based on YOLO v3 framework and lightweight neural network MobileNet v3. Compared with YOLO v3, the size of the improved network is reduced by 2.9 times at the same level of detection precision. Experimental results showed that the detection speed of the improved lightweight network could reach up to 40.35FPS, with the mean average precision (mAP) of 87.8%.
DOI:10.1109/ICCECE51280.2021.9342470