Small Target UAV Detection Algorithm Based on Dynamic Partial Residual Network and Context Guidance.

Saved in:

Bibliographic Details
Title:	Small Target UAV Detection Algorithm Based on Dynamic Partial Residual Network and Context Guidance.
Authors:	Zhang, Bo¹ zber@163.com, Liang, Jie² 2824586575@qq.com
Source:	Engineering Letters. Jul2026, Vol. 34 Issue 7, p2534-2546. 13p.
Subjects:	Drone surveillance, Feature extraction, Object recognition (Computer vision), Artificial neural networks
Abstract:	To address the problems of low accuracy and severe missed detections in UAV detection caused by small target scale and complex and variable backgrounds, while reducing the number of model parameters and computational load, a multi-scale partial residual feature fusion network is proposed. First, to address the difficulty of UAV feature extraction caused by the complex and ever-changing flight background of UAVs, a dynamic partial residual feature extraction module is proposed. This module conducts partial convolution on the channels of the feature map and then interacts and fuses with the non-convolved part. Reduce the complexity of the model and enhance its feature extraction capabilities. Secondly, an Adaptive Spatial Feature Reconstruction Pyramid Network is introduced. This network captures the global context in both the horizontal and vertical directions, enabling the model to pay more attention to the key feature information areas and strengthening the model's multi-scale feature fusion ability. Finally, by combining the dynamic focusing mechanism with the Gaussian distribution, dynamic sample balancing is achieved, thereby improving the detection accuracy of small-target UAVs. The experimental results on the DUT-Anti-UAV dataset demonstrate that, compared with the original model, the false negative rate (FNR) is lowered by 2.1%, while the parameter count and computational complexity (GFLOPS) are decreased by 15.4% and 14.4%, respectively. Meanwhile, the precision, recall, and mAP@0.5 were improved by 1.7%, 2.1%, and 1.9%, respectively, compared to the original model. These results validate the superior performance of the improved model in detecting small-target UAVs. [ABSTRACT FROM AUTHOR]
	Copyright of Engineering Letters is the property of International Association of Engineers (IAENG) and its content may not be copied or emailed to multiple sites without the copyright holder's express written permission. Additionally, content may not be used with any artificial intelligence tools or machine learning technologies. However, users may print, download, or email articles for individual use. This abstract may be abridged. No warranty is given about the accuracy of the copy. Users should refer to the original published version of the material for the full abstract. (Copyright applies to all Abstracts.)
Database:	Engineering Source

Description
Abstract:	To address the problems of low accuracy and severe missed detections in UAV detection caused by small target scale and complex and variable backgrounds, while reducing the number of model parameters and computational load, a multi-scale partial residual feature fusion network is proposed. First, to address the difficulty of UAV feature extraction caused by the complex and ever-changing flight background of UAVs, a dynamic partial residual feature extraction module is proposed. This module conducts partial convolution on the channels of the feature map and then interacts and fuses with the non-convolved part. Reduce the complexity of the model and enhance its feature extraction capabilities. Secondly, an Adaptive Spatial Feature Reconstruction Pyramid Network is introduced. This network captures the global context in both the horizontal and vertical directions, enabling the model to pay more attention to the key feature information areas and strengthening the model's multi-scale feature fusion ability. Finally, by combining the dynamic focusing mechanism with the Gaussian distribution, dynamic sample balancing is achieved, thereby improving the detection accuracy of small-target UAVs. The experimental results on the DUT-Anti-UAV dataset demonstrate that, compared with the original model, the false negative rate (FNR) is lowered by 2.1%, while the parameter count and computational complexity (GFLOPS) are decreased by 15.4% and 14.4%, respectively. Meanwhile, the precision, recall, and mAP@0.5 were improved by 1.7%, 2.1%, and 1.9%, respectively, compared to the original model. These results validate the superior performance of the improved model in detecting small-target UAVs. [ABSTRACT FROM AUTHOR]
ISSN:	1816093X