针对复杂场景下小地物漏检,裸地、道路等干扰因素引起建筑物变化样本不足导致的精度低等问题,提出一种改进的金字塔场景解析网络(Pyramid Scene Parsing Network,PSPNet)。搭建孪生PSPNet网络作为基础模型,通过融合不同尺度的特征层信息...针对复杂场景下小地物漏检,裸地、道路等干扰因素引起建筑物变化样本不足导致的精度低等问题,提出一种改进的金字塔场景解析网络(Pyramid Scene Parsing Network,PSPNet)。搭建孪生PSPNet网络作为基础模型,通过融合不同尺度的特征层信息,提高小尺度地物的检测精度;其次针对变化样本不足问题,结合多任务思想,使语义任务和变化检测任务在同一个网络中进行,从而解决变化样本不足的问题。结果表明:改进后的PSPNet模型在建筑物变化检测中的精确率为92.35%,召回率为85.61%,F_(1)分数为0.8885。相比原始的PSPNet模型精度提升6.2%、12.03%和0.09。本研究可为复杂场景下建筑物变化检测提供技术支持。展开更多
针对高分辨率遥感影像中建筑目标较小和背景信息冗余带来的挑战,提出了一种称为FE-DETR(feature enhancement-detection with transformer)的端到端目标检测算法。首先,利用拼接融合模块(concatenation fusion module,CFM)融合不同尺度...针对高分辨率遥感影像中建筑目标较小和背景信息冗余带来的挑战,提出了一种称为FE-DETR(feature enhancement-detection with transformer)的端到端目标检测算法。首先,利用拼接融合模块(concatenation fusion module,CFM)融合不同尺度的特征层,缓解小建筑目标特征缺失问题;其次,使用全局通道注意力(global channel attention,GCA)模块细化融合后的特征。具体来说,该模块通过构建通道间的关系矩阵,提高模型对目标的感知能力,有效缓解复杂背景信息带来的干扰。最后,在WCH(Wuhan caidian house)、EA(east Asia)和CBC(city building of China)数据集上评估该算法的检测性能。实验结果表明,所提出的改进算法在上述3个数据集上AP_(50)分别提高了0.8%、0.6%和0.6%,验证了该算法的有效性。展开更多
This paper presents a novel method, which enhances the use of external mechanisms by considering a multisensor system, composed of sonars and a CCD camera. Monocular vision provides redundant information about the loc...This paper presents a novel method, which enhances the use of external mechanisms by considering a multisensor system, composed of sonars and a CCD camera. Monocular vision provides redundant information about the location of the geometric entities detected by the sonar sensors. To reduce ambiguity significantly, an improved and more detailed sonar model is utilized. Moreover, Hough transform is used to extract features from raw sonar data and vision image. Information is fused at the level of features. This technique significantly improves the reliability and precision of the environment observations used for the simultaneous localization and map building problem for mobile robots. Experimental results validate the favorable performance of this approach.展开更多
文摘针对复杂场景下小地物漏检,裸地、道路等干扰因素引起建筑物变化样本不足导致的精度低等问题,提出一种改进的金字塔场景解析网络(Pyramid Scene Parsing Network,PSPNet)。搭建孪生PSPNet网络作为基础模型,通过融合不同尺度的特征层信息,提高小尺度地物的检测精度;其次针对变化样本不足问题,结合多任务思想,使语义任务和变化检测任务在同一个网络中进行,从而解决变化样本不足的问题。结果表明:改进后的PSPNet模型在建筑物变化检测中的精确率为92.35%,召回率为85.61%,F_(1)分数为0.8885。相比原始的PSPNet模型精度提升6.2%、12.03%和0.09。本研究可为复杂场景下建筑物变化检测提供技术支持。
文摘针对高分辨率遥感影像中建筑目标较小和背景信息冗余带来的挑战,提出了一种称为FE-DETR(feature enhancement-detection with transformer)的端到端目标检测算法。首先,利用拼接融合模块(concatenation fusion module,CFM)融合不同尺度的特征层,缓解小建筑目标特征缺失问题;其次,使用全局通道注意力(global channel attention,GCA)模块细化融合后的特征。具体来说,该模块通过构建通道间的关系矩阵,提高模型对目标的感知能力,有效缓解复杂背景信息带来的干扰。最后,在WCH(Wuhan caidian house)、EA(east Asia)和CBC(city building of China)数据集上评估该算法的检测性能。实验结果表明,所提出的改进算法在上述3个数据集上AP_(50)分别提高了0.8%、0.6%和0.6%,验证了该算法的有效性。
基金supported by the National Natural Science Foundation of China (No.60805032)the National High Technology Research and Development Program (No.2006AA040202, 2007AA041703)
文摘This paper presents a novel method, which enhances the use of external mechanisms by considering a multisensor system, composed of sonars and a CCD camera. Monocular vision provides redundant information about the location of the geometric entities detected by the sonar sensors. To reduce ambiguity significantly, an improved and more detailed sonar model is utilized. Moreover, Hough transform is used to extract features from raw sonar data and vision image. Information is fused at the level of features. This technique significantly improves the reliability and precision of the environment observations used for the simultaneous localization and map building problem for mobile robots. Experimental results validate the favorable performance of this approach.
