Research on remote sensing small target detection algorithm based on context refined feature network
DONG Yan
HE Yuan
LI Bingrui
GAO Guangshuai
WEI Minghong
Abstract：In remote sensing images,small targets,due to their limited size and complex back-grounds,often suffer from the loss of critical information or the submergence of their features by background characteristics after multiple layers of convolutional processing.This increases the diffi-culty of subsequent classification and regression tasks.To address this issue,this paper proposes a re-mote sensing small target detection algorithm based on a refined contextual feature network.The al-gorithm first adjusts the distribution of the number of Bottlenecks in the backbone network to pre-serve more detailed information from the lower layers.Secondly,it introduces an enhanced spatial at-tention module on top of the feature pyramid to generate hierarchical attention heatmaps that highlight features of specific scales at different levels,and combines this with a feature guidance module to in-tegrate contextual information.Finally,an adaptive sparse convolution module is introduced to reduce the risk of model overfitting and decrease computational load.Experimental results on the TinyPerson dataset show that the algorithm achieves an average precision of 58.24％when the IoU threshold is set to 0.5,outperforming existing detection algorithms.
Keywords：remote sensing target detectionstrengthen spatial attentionhierarchical attentiona-daptive sparse convolution
Publication Date：2025-04-25
Online Publishing Date：2025-08-15(First online date of this platform, not the publication date of the document)
Pages：10( 15-23,33 )