Urban object detection algorithm based on feature enhancement and progressive dynamic aggregation strategy

Abstract

Traditional target detection models face challenges in recognizing urban high-altitude remote sensing targets due to complex background noise and significant variations in target scale. These challenges can result in loss of feature information and missed object detection. In light of this, this article introduces a novel dual-gated feature mechanism and adaptive fusion strategy. First, the dual-gated feature mechanism enables selective suppression or enhancement of multilevel features, thereby reducing the interference of complex environmental noise in remote sensing on feature fusion. Second, the adaptive fusion strategy and module facilitate multilevel scale feature fusion, and by dynamically learning fusion weights, they mitigate scale conflicts during the feature extraction process and preserve feature information. Experimental comparisons and analysis on the RSOD and NWPU VHR-10 public datasets showcase the effectiveness of the proposed method. In comparison to current mainstream detection methods, the improved approach presented in this article demonstrates significant advantages in terms of detection performance and efficiency.

Keywords:

• Remote sensing
• algorithm feature extraction
• image fusion

Acknowledgements

We would like to express our gratitude for the encouragement and support from our co-authors. The ‘DIOR’ large-scale benchmark dataset proposed by Han Junwei’s team at Xi’an University of Technology was an extremely important resource that provided valuable information. We also thank MathWorks for providing the excellent MATLAB software package, which is widely available and comes with documentation that helped us complete this work under the Windows operating system.

Disclosure statement

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this article.

Authors’ contributions

Luxuan Bian performed the data analysis; Bo Li performed the formal analysis; Jue Wang performed the validation; Zijun Gao wrote the manuscript.

Data availability statement

The NWPU VHR-10 dataset is freely available as follows, https://github.com/RSIA-LIESMARS-WHU/RSOD-Dataset. The RSOD dataset are freely available as follows: http://www.escience.cn/people/gongcheng/DIOR.html. And the relabelled images and codes that support the findings of this study are available from the corresponding author, upon reasonable request.