对于视觉场景的理解是机器人在未知环境中进行有目的的行动的一项重要能力,图像语义分割能够有效地帮助机器人理解周围场景的语义特性。条件随机场(Conditional Random Field,CRF)是解决语义分割问题的一个重要框架。针对传统条件随机...对于视觉场景的理解是机器人在未知环境中进行有目的的行动的一项重要能力,图像语义分割能够有效地帮助机器人理解周围场景的语义特性。条件随机场(Conditional Random Field,CRF)是解决语义分割问题的一个重要框架。针对传统条件随机场相邻节点数过于稀疏的局限性,研究了具有稀疏高阶势CRF的图像语义分割算法,提出一种高阶CRF的二次规划(quadratic programming,QP)松弛推理算法。首先,使用来自TextonBoost的一元势、高斯二元势以及由Pn-Potts模型推导得到的高阶势建立能量函数,然后利用高效的QP松弛推理算法来解决高阶CRF的能量最小化问题,最后采用Pascal VOC2012公开数据集进行实验,验证算法的可行性与有效性。实验证明,该算法有效地克服了传统条件随机场局限性,获得了更好的语义分割结果。展开更多
Global climate change has been found to substantially influence the phenology of rangeland,especially on the Tibetan Plateau. However, there is considerable controversy about the trends and causes of rangeland phenolo...Global climate change has been found to substantially influence the phenology of rangeland,especially on the Tibetan Plateau. However, there is considerable controversy about the trends and causes of rangeland phenology owing to different phenological exploration methods and lack of ground validation. Little is known about the uncertainty in the exploration accuracy of vegetation phenology.Therefore, in this study, we selected a typical alpine rangeland near Damxung national meteorological station as a case study on central Tibetan Plateau, and identified several important sources influencing phenology to better understand their effects on phenological exploration. We found man-made land use was not easily distinguished from natural rangelands, and therefore this may confound phenological response to climate change in the rangeland. Change trends of phenology explored by four methods were similar, but ratio threshold method(RTM) was more suitable for exploring vegetation phenology in terms of the beginning of growing season(BGS) and end of growing season(EGS). However, some adjustments are needed when RTM is used in extreme drought years. MODIS NDVI/EVI dataset was most suitable for exploring vegetation phenology of BGS and EGS. The discrimination capacities of vegetation phenology declined with decreasing resolution of remote sensing images from MODIS to GIMMS AVHRR datasets. Additionally, distinct trends of phenological change rates were indicated in different terrain conditions, with advance of growing season in high altitudes but delay of season in lower altitudes. Therefore, it was necessary to eliminate interference of complex terrain and man-made land use to ensure the representativeness of natural vegetation. Moreover, selecting the appropriate method to explore rangelands and fully considering the impact of topography are important to accurately analyze the effects of climate change on vegetation phenology.展开更多
文摘对于视觉场景的理解是机器人在未知环境中进行有目的的行动的一项重要能力,图像语义分割能够有效地帮助机器人理解周围场景的语义特性。条件随机场(Conditional Random Field,CRF)是解决语义分割问题的一个重要框架。针对传统条件随机场相邻节点数过于稀疏的局限性,研究了具有稀疏高阶势CRF的图像语义分割算法,提出一种高阶CRF的二次规划(quadratic programming,QP)松弛推理算法。首先,使用来自TextonBoost的一元势、高斯二元势以及由Pn-Potts模型推导得到的高阶势建立能量函数,然后利用高效的QP松弛推理算法来解决高阶CRF的能量最小化问题,最后采用Pascal VOC2012公开数据集进行实验,验证算法的可行性与有效性。实验证明,该算法有效地克服了传统条件随机场局限性,获得了更好的语义分割结果。
基金supported by the National Natural Science Foundation of China (41271067)National key research and development program (2016YFC0502001)
文摘Global climate change has been found to substantially influence the phenology of rangeland,especially on the Tibetan Plateau. However, there is considerable controversy about the trends and causes of rangeland phenology owing to different phenological exploration methods and lack of ground validation. Little is known about the uncertainty in the exploration accuracy of vegetation phenology.Therefore, in this study, we selected a typical alpine rangeland near Damxung national meteorological station as a case study on central Tibetan Plateau, and identified several important sources influencing phenology to better understand their effects on phenological exploration. We found man-made land use was not easily distinguished from natural rangelands, and therefore this may confound phenological response to climate change in the rangeland. Change trends of phenology explored by four methods were similar, but ratio threshold method(RTM) was more suitable for exploring vegetation phenology in terms of the beginning of growing season(BGS) and end of growing season(EGS). However, some adjustments are needed when RTM is used in extreme drought years. MODIS NDVI/EVI dataset was most suitable for exploring vegetation phenology of BGS and EGS. The discrimination capacities of vegetation phenology declined with decreasing resolution of remote sensing images from MODIS to GIMMS AVHRR datasets. Additionally, distinct trends of phenological change rates were indicated in different terrain conditions, with advance of growing season in high altitudes but delay of season in lower altitudes. Therefore, it was necessary to eliminate interference of complex terrain and man-made land use to ensure the representativeness of natural vegetation. Moreover, selecting the appropriate method to explore rangelands and fully considering the impact of topography are important to accurately analyze the effects of climate change on vegetation phenology.
