融合多尺度边缘增强提取的YOLOv8遥感图像目标检测算法OA

In order to enhance the feature extraction capability and detection accuracy of YOLOv8 algorithm for targets of diverse scales in remote sensing tasks,a multi-scale edge enhancement extraction YOLOv8 algo-rithm is proposed.Shallow robust efficient multi-scale downsampling and deep robust efficient multi-scale downsampling modules are introduced to enhance the ability to preserve low and deep level feature details,re-spectively.In addition,an efficient edge enhanced upsampling module is introduced to improve the network's detection capability under multi-scale and complex background conditions.Furthermore,a partial self-attention mechanism module is integrated to enhance global information modeling capabilities and effectively suppress background noise.Experimental results show that,compared to the original YOLOv8 algorithm,the proposed algorithm achieves superior performance on the DIOR dataset,with an accuracy improvement of 0.7%,a re-call improvement of 2.4%,an average accuracy mean value improvement of 2.0%for the intersection-over-union of 0.50,and an average accuracy mean value improvement of 2.8%for the intersection-over-union of 0.50 to 0.95.