矿井人员图像分割是实现煤矿井下人员检测、行为识别、视频定位跟踪等技术的基础任务之一。然而,由于矿井下环境特殊,常规图像分割方法均难以满足对井下人员的精准分割要求。为解决矿井人员图像的分割问题,提出一种基于超像素粒化及同...矿井人员图像分割是实现煤矿井下人员检测、行为识别、视频定位跟踪等技术的基础任务之一。然而,由于矿井下环境特殊,常规图像分割方法均难以满足对井下人员的精准分割要求。为解决矿井人员图像的分割问题,提出一种基于超像素粒化及同质图像粒聚类的分割方法,能够适用于煤矿井下多种场景的人员图像。首先,使用简单线性迭代聚类(Simple Linear Iterative Clustering,SLIC)模型将井下人员图像初始粒化为超像素单元,并通过测量离线样本图像中所标记人员像素点与超像素之间的RGB相似度值判定人员超像素。其次,由邻居超像素辅助检测欠分割人员超像素并将其彻底分割为2个子超像素单元,选择其中之一的精英人员超像素并提取其纹理和灰度特征。接着,将具有最相似图像特征的邻接精英人员超像素定义为同质图像粒,同质图像粒相互融合并聚类形成具有特定语义信息的同质人员区域。最后,由所有同质人员区域共同构成完整的人员区域,并实现人员区域与图像背景的分离。通过对煤矿井下4种场景下的人员图像进行算法性能验证,实验结果表明:超像素粒化算法的F-Measure值分别较对比算法平均值高出2.11%,3.36%,13.16%,6.82%,同质人员图像粒聚类算法精度值分别达到99.0%,100%,94.4%和93.75%,并且所提分割方法对井下4种不同场景中的人员图像均具有较强的鲁棒性和较好的分割效果。展开更多
在纹理丰富的高光谱图像中获得精确的噪声估计,是噪声估计任务中的难点。本文基于高光谱图像的空间规律性和光谱相关性,提出一种基于超像素分割的光谱去相关法。同质区域划分是许多噪声估计方法的关键步骤,精确的同质区域划分能有效提...在纹理丰富的高光谱图像中获得精确的噪声估计,是噪声估计任务中的难点。本文基于高光谱图像的空间规律性和光谱相关性,提出一种基于超像素分割的光谱去相关法。同质区域划分是许多噪声估计方法的关键步骤,精确的同质区域划分能有效提高噪声估计精度。为此,将简单线性迭代聚类算法(Simple linear iterative clustering algorithm,SLIC)与光谱-空间相似性结合,划分高光谱图像为局部结构相似的图像块,以保持同质特征;为了提高光谱间的区分能力,将光谱信息散度和光谱角联合作为光谱距离;结合多元线性回归在同质区域内去除光谱相关性,在获得的残差图上估计噪声水平。对不同地物复杂程度的模拟图像,添加不同程度的噪声,通过与多种方法比较,验证了本文方法的有效性和稳定性。最后,本文方法成功应用于Urban数据的噪声水平估计,准确识别出受噪声严重污染的波段。展开更多
The macroscopically-zoned grandite from the garnetite skarn of Meka Presedla （Kopaonik Mountain, Serbia） was studied with optical microscopy, electron microprobe analysis （EMPA）, Fourier transform infra-red （FT-I...The macroscopically-zoned grandite from the garnetite skarn of Meka Presedla （Kopaonik Mountain, Serbia） was studied with optical microscopy, electron microprobe analysis （EMPA）, Fourier transform infra-red （FT-IR）, and Raman methods. The EMPA results indicate that the main core-rim compositional variations （Ca2.93-2.97Mn0.05-0.06Mg0.00-0.01AI1.14-L26 Fe0.72-0.83 Ti0.00-0.02Si2.97-3.02012） slightly differ along the zones, showing evidence for a quasi-cyclic alternation of the oscillatory zoning nature. Among this, considerable variation is observed only by the AI-Fe3＋ substitutions in the octahedral site. The EMPA also indicate that the grandite zones compositionally vary, mostly within ＋1 and ±2 mol% of the homogeneity level range, that is, Grs64±1Adra36±1SpS2 （A）, Grs62±1Adra38±1Sps2 （B）, Grs59±2Adr40±2Sps2 （C）, Grs58±2Adr41±2Sps2 （D）, and Grss±1Adr41±1Sps2 （E）. Therefore, the investigated garnet can be considered as relatively highly homogeneous. The majority of compositions lie within the narrow miscibility region of 0.58±2〈XGrs〈0.64±1, without gaps, and with only three outliers near the zone boundaries of approximately 0.38〈XGrs〈0.52. FT-IR and Raman bands are almost constant within the zones and adequate to the chemical compositions. All of the zones should be considered as anhydrous. From the results, formation temperatures between -600 and 720℃ and pressures 0f-2-3 kbars, are derived. Among five possible causes for the slightly optical anisotropy of grandite, three were reconsidered, and consequently rejected.展开更多
文摘矿井人员图像分割是实现煤矿井下人员检测、行为识别、视频定位跟踪等技术的基础任务之一。然而,由于矿井下环境特殊,常规图像分割方法均难以满足对井下人员的精准分割要求。为解决矿井人员图像的分割问题,提出一种基于超像素粒化及同质图像粒聚类的分割方法,能够适用于煤矿井下多种场景的人员图像。首先,使用简单线性迭代聚类(Simple Linear Iterative Clustering,SLIC)模型将井下人员图像初始粒化为超像素单元,并通过测量离线样本图像中所标记人员像素点与超像素之间的RGB相似度值判定人员超像素。其次,由邻居超像素辅助检测欠分割人员超像素并将其彻底分割为2个子超像素单元,选择其中之一的精英人员超像素并提取其纹理和灰度特征。接着,将具有最相似图像特征的邻接精英人员超像素定义为同质图像粒,同质图像粒相互融合并聚类形成具有特定语义信息的同质人员区域。最后,由所有同质人员区域共同构成完整的人员区域,并实现人员区域与图像背景的分离。通过对煤矿井下4种场景下的人员图像进行算法性能验证,实验结果表明:超像素粒化算法的F-Measure值分别较对比算法平均值高出2.11%,3.36%,13.16%,6.82%,同质人员图像粒聚类算法精度值分别达到99.0%,100%,94.4%和93.75%,并且所提分割方法对井下4种不同场景中的人员图像均具有较强的鲁棒性和较好的分割效果。
文摘在纹理丰富的高光谱图像中获得精确的噪声估计,是噪声估计任务中的难点。本文基于高光谱图像的空间规律性和光谱相关性,提出一种基于超像素分割的光谱去相关法。同质区域划分是许多噪声估计方法的关键步骤,精确的同质区域划分能有效提高噪声估计精度。为此,将简单线性迭代聚类算法(Simple linear iterative clustering algorithm,SLIC)与光谱-空间相似性结合,划分高光谱图像为局部结构相似的图像块,以保持同质特征;为了提高光谱间的区分能力,将光谱信息散度和光谱角联合作为光谱距离;结合多元线性回归在同质区域内去除光谱相关性,在获得的残差图上估计噪声水平。对不同地物复杂程度的模拟图像,添加不同程度的噪声,通过与多种方法比较,验证了本文方法的有效性和稳定性。最后,本文方法成功应用于Urban数据的噪声水平估计,准确识别出受噪声严重污染的波段。
基金supported by the Serbian Ministry of Science and Environmental Protection (projectno. 1992 and 142055)
文摘The macroscopically-zoned grandite from the garnetite skarn of Meka Presedla （Kopaonik Mountain, Serbia） was studied with optical microscopy, electron microprobe analysis （EMPA）, Fourier transform infra-red （FT-IR）, and Raman methods. The EMPA results indicate that the main core-rim compositional variations （Ca2.93-2.97Mn0.05-0.06Mg0.00-0.01AI1.14-L26 Fe0.72-0.83 Ti0.00-0.02Si2.97-3.02012） slightly differ along the zones, showing evidence for a quasi-cyclic alternation of the oscillatory zoning nature. Among this, considerable variation is observed only by the AI-Fe3＋ substitutions in the octahedral site. The EMPA also indicate that the grandite zones compositionally vary, mostly within ＋1 and ±2 mol% of the homogeneity level range, that is, Grs64±1Adra36±1SpS2 （A）, Grs62±1Adra38±1Sps2 （B）, Grs59±2Adr40±2Sps2 （C）, Grs58±2Adr41±2Sps2 （D）, and Grss±1Adr41±1Sps2 （E）. Therefore, the investigated garnet can be considered as relatively highly homogeneous. The majority of compositions lie within the narrow miscibility region of 0.58±2〈XGrs〈0.64±1, without gaps, and with only three outliers near the zone boundaries of approximately 0.38〈XGrs〈0.52. FT-IR and Raman bands are almost constant within the zones and adequate to the chemical compositions. All of the zones should be considered as anhydrous. From the results, formation temperatures between -600 and 720℃ and pressures 0f-2-3 kbars, are derived. Among five possible causes for the slightly optical anisotropy of grandite, three were reconsidered, and consequently rejected.
