基于边缘增强和多尺度特征融合的遥感图像船舰检测OA
Ship detection in remote sensing images using edge enhancement and multi-scale feature fusion
遥感图像中的船舰目标具有尺度变化剧烈、分布密集和方向朝向任意的特点,特别是船舰与海洋环境之间对比度低,相邻船舰之间边界模糊,这使船舰检测面临更高的挑战.针对以上问题,本文提出了一种基于边缘增强和多尺度特征融合的遥感图像船舰检测模型.设计了高频特征增强模块,提升模型捕获细节的能力;提出了一种边缘信息引导的多尺度特征融合方法,缓解浅层边缘信息在传递过程中丢失的问题;构建轻量化定向检测头,减少模型参数量.实验结果表明,改进后的模型在ShipRSImageNet数据集和HRSC2016数据集上,平均检测精度(mAP50)较YOLO11-obb模型分别提升3.6和2.1百分点,有效提升遥感图像船舰检测的精度.
Ship objects in remote sensing images exhibit large scale variation,dense distribution,and arbitrary orienta-tion.In particular,the low contrast between ships and the ocean background,along with blurred boundaries between ad-jacent ships,poses greater challenges for detection.To address these issues,a model based on edge enhancement and multi-scale feature fusion for ship detection in remote sensing images was proposed.Firstly,a high-frequency feature enhancement module was designed to improve the ability of the model to capture fine details.Furthermore,an edge-guided multi-scale feature fusion method was proposed to mitigate the loss of edge information on low-level during propagation.Finally,a lightweight oriented detection head was constructed to reduce the params of the model Experi-mental results show that the improved model improves 3.6 and 2.1 percentage points of mAP50 on the ShipRSImageN-et dataset and the HRSC2016 Dataset,compared to the YOLO1 l-obb model,effectively improves the accuracy of ship detection in remote sensing images.
王德文;宋学帅;李成浩;赵文清
华北电力大学计算机系,河北保定 071003||河北省能源电力知识计算重点实验室,河北保定 071003华北电力大学计算机系,河北保定 071003华北电力大学计算机系,河北保定 071003华北电力大学计算机系,河北保定 071003||复杂能源系统智能计算教育部工程研究中心,河北保定 071003
信息技术与安全科学
遥感图像船舰检测高频特征边缘增强多尺度特征融合轻量化检测头YOLO
remote sensing imageship detectionhigh-frequency featureedge enhancementmulti-scalefeature fu-sionlightweight detection headYOLO
《智能系统学报》 2026 (1)
60-71,12
国家自然科学基金项目(62371188).
评论