SFTN: Fast object detection for aerial images

Autor: Li Chen, Fan Zhang, Wei Guo, Tianyang Li, Mingqian Sun
Jazyk: angličtina
Rok vydání: 2023
Předmět:
Zdroj: IET Image Processing, Vol 17, Iss 13, Pp 3897-3907 (2023)
Druh dokumentu: article
ISSN: 1751-9667
1751-9659
DOI: 10.1049/ipr2.12906
Popis: Abstract The task of remote sensing image object detection in low latency scenes is of great research significance. To address the problem that the current high‐precision object detection algorithm based on a feature pyramid network is slow due to a large number of parameters and complicated computation, a fast remote sensing image object detection method based on a Single‐scale Feature Transformation Network (SFTN) is proposed. Firstly, based on the single‐scale remote sensing image features, the new channel features are quickly generated by a linear transformation of the original features and convolution kernel clustering optimization using cosine similarity; secondly, in order to obtain multi‐scale receptive fields, a parallel residual hole convolution module is designed to cover multi‐category remote sensing object scales on the feature map; finally, angle variables are introduced and optimized using angle similarity to effectively improve the object orientation accuracy. The experimental results on different datasets show that the method in this paper improves the detection speed rapidly while ensuring the accuracy of remote sensing image object detection, which is better than many remote sensing image object detection methods. The results demonstrate the reliability and robustness of the method.
Databáze: Directory of Open Access Journals
Externí odkaz: