翻唱歌曲识别是音乐信息检索(Music Information Retrieval, MIR)领域最具挑战性的任务之一。为了提高翻唱歌曲识别(Cover Song Identification, CSI)的准确率,研究者提出了多音频特征相似度张量积图融合的翻唱识别模型,但相似度高维空...翻唱歌曲识别是音乐信息检索(Music Information Retrieval, MIR)领域最具挑战性的任务之一。为了提高翻唱歌曲识别(Cover Song Identification, CSI)的准确率,研究者提出了多音频特征相似度张量积图融合的翻唱识别模型,但相似度高维空间几何结构的学习大幅度增加了模型的时间复杂度,同时该模型没有考虑歌词对翻唱识别的重要性。本文提出了基于音频内容和歌词文本相似度融合的翻唱识别模型。采用深度学习的方法分别提取音频特征和歌词特征,并采用相似度网络融合模型对这两种特征的相似度进行融合。为了验证算法的有效性,构建了Covers2326多模态数据库。实验结果表明,与基于多音频特征相似度张量积图融合模型相比,本文模型取得了更高的识别准确率和更低的时间复杂度。展开更多
Based on data on taxed-cropland area and on the number of households in historical documents, a probabilistic model of cropland distribution and a cropland area allocation model were designed and validated. Cropland a...Based on data on taxed-cropland area and on the number of households in historical documents, a probabilistic model of cropland distribution and a cropland area allocation model were designed and validated. Cropland areas for the years AD976, 997, 1066, and 1078 were estimated at the level of Lu（an administrative region of the Northern Song Dynasty）. The results indicated that（1） the cropland area of the whole study region for AD976, 997, 1066, and 1078 was about 468.27 million mu（a Chinese unit of area, with1 mu=666.7m2）, 495.53 million mu, 697.65 million mu, and 731.94 million mu, respectively. The fractional cropland area（FCA） increased from 10.7% to 16.8%, and the per capita cropland area decreased from 15.7 mu to 8.4 mu.（2） With regard to the cropland spatial pattern, the FCA of the southeast, north, and southwest regions of the Northern Song territory increased by 12.0%, 5.2%, and 1.2%, respectively. The FCA of some regions in the Yangtze River Plain increased to greater than 40%, and the FCA of the North China Plain increased to greater than 20%. However, the FCA of the southwest region（except for the Chengdu Plain） in the Northern Song territory was less than 6%.（3） There were 84.2% Lus whose absolute relative error was smaller than 20% in the mid Northern Song Dynasty. The validation results indicate that our models are reasonable and that the results of reconstruction are credible.展开更多
To understand historical human-induced land cover change and its climatic effects, it is necessary to create historical land use datasets with explicit spatial information. Using the taxes-cropland area and number of ...To understand historical human-induced land cover change and its climatic effects, it is necessary to create historical land use datasets with explicit spatial information. Using the taxes-cropland area and number of families compiled from historical documents, we estimated the real cropland area and populations within each Lu （a province-level political region in the Northern Song Dynasty） in the mid-Northern Song Dynasty （AD1004-1085）. The estimations were accomplished through analyzing the contemporary policies of tax, population and agricultural development. Then, we converted the political region-based cropland area to geographically explicit grid cell-based fractional cropland at the cell size of 60 km by 60 km. The conversion was based on calculating cultivation suitability of each grid cell using the topographic slope, altitude and population density as the independent variables. As a result, the total area of cropland within the Northern Song territory in the 1070s was estimated to be about 720 million mu （Chinese area unit, 1 mu = 666.7 m2）, of which 40.1% and 59.9% oc- curred in the north and south respectively. The population was estimated to be about 87.2 million, of which 38.7% and 61.3% were in the north and south respectively, and per capita cropland area was about 8.2 mu. The national mean reclamation ratio （i.e. ratio of cropland area to total land area; RRA hereafter for short） was bout 16.6%. The plain areas, such as the North China Plain, the middle and lower reaches of the Yangtze River, Guanzhong Plain, plains surrounding the Dongting Lake and Poyang Lake and Sichuan Basin, had a higher RRA, being mostly over 40%; while the hilly and mountainous areas, such as south of Nanling Mountains, the southwest regions （excluding the Chengdu Plain）, Loess Plateau and south- east coastal regions, had a lower RRA, being less than 20%. Moreover, RRA varied with topographic slope and altitude. In the areas of low altitude （~〈250 m）, middle altitude （250-100 m） and high altitude （1000-3500 m）, there were 443 million, 215 million and 64 million mu of cropland respectively and their regional mean RRAs were 27.5%, 12.6% and 7.2% respectively. In the areas of flat slope, gentle slope, medium slope and steep slope, there were 116 million, 456 million, 144 million and 2 million mu of cropland respectively and their regional mean RRAs were 34.6%, 20.7%, 8.5% and 2.3% respectively.展开更多
文摘翻唱歌曲识别是音乐信息检索(Music Information Retrieval, MIR)领域最具挑战性的任务之一。为了提高翻唱歌曲识别(Cover Song Identification, CSI)的准确率,研究者提出了多音频特征相似度张量积图融合的翻唱识别模型,但相似度高维空间几何结构的学习大幅度增加了模型的时间复杂度,同时该模型没有考虑歌词对翻唱识别的重要性。本文提出了基于音频内容和歌词文本相似度融合的翻唱识别模型。采用深度学习的方法分别提取音频特征和歌词特征,并采用相似度网络融合模型对这两种特征的相似度进行融合。为了验证算法的有效性,构建了Covers2326多模态数据库。实验结果表明,与基于多音频特征相似度张量积图融合模型相比,本文模型取得了更高的识别准确率和更低的时间复杂度。
基金National Natural Science Foundation of China,No.41271227 The Special Program for Basic Work of the Ministry of Science and Technology,China,No.2014FY210900
文摘Based on data on taxed-cropland area and on the number of households in historical documents, a probabilistic model of cropland distribution and a cropland area allocation model were designed and validated. Cropland areas for the years AD976, 997, 1066, and 1078 were estimated at the level of Lu（an administrative region of the Northern Song Dynasty）. The results indicated that（1） the cropland area of the whole study region for AD976, 997, 1066, and 1078 was about 468.27 million mu（a Chinese unit of area, with1 mu=666.7m2）, 495.53 million mu, 697.65 million mu, and 731.94 million mu, respectively. The fractional cropland area（FCA） increased from 10.7% to 16.8%, and the per capita cropland area decreased from 15.7 mu to 8.4 mu.（2） With regard to the cropland spatial pattern, the FCA of the southeast, north, and southwest regions of the Northern Song territory increased by 12.0%, 5.2%, and 1.2%, respectively. The FCA of some regions in the Yangtze River Plain increased to greater than 40%, and the FCA of the North China Plain increased to greater than 20%. However, the FCA of the southwest region（except for the Chengdu Plain） in the Northern Song territory was less than 6%.（3） There were 84.2% Lus whose absolute relative error was smaller than 20% in the mid Northern Song Dynasty. The validation results indicate that our models are reasonable and that the results of reconstruction are credible.
基金China Global Change Research Program,No.2010CB950102No.2010CB950901National Natural Science Foundation of China,No.40971061
文摘To understand historical human-induced land cover change and its climatic effects, it is necessary to create historical land use datasets with explicit spatial information. Using the taxes-cropland area and number of families compiled from historical documents, we estimated the real cropland area and populations within each Lu （a province-level political region in the Northern Song Dynasty） in the mid-Northern Song Dynasty （AD1004-1085）. The estimations were accomplished through analyzing the contemporary policies of tax, population and agricultural development. Then, we converted the political region-based cropland area to geographically explicit grid cell-based fractional cropland at the cell size of 60 km by 60 km. The conversion was based on calculating cultivation suitability of each grid cell using the topographic slope, altitude and population density as the independent variables. As a result, the total area of cropland within the Northern Song territory in the 1070s was estimated to be about 720 million mu （Chinese area unit, 1 mu = 666.7 m2）, of which 40.1% and 59.9% oc- curred in the north and south respectively. The population was estimated to be about 87.2 million, of which 38.7% and 61.3% were in the north and south respectively, and per capita cropland area was about 8.2 mu. The national mean reclamation ratio （i.e. ratio of cropland area to total land area; RRA hereafter for short） was bout 16.6%. The plain areas, such as the North China Plain, the middle and lower reaches of the Yangtze River, Guanzhong Plain, plains surrounding the Dongting Lake and Poyang Lake and Sichuan Basin, had a higher RRA, being mostly over 40%; while the hilly and mountainous areas, such as south of Nanling Mountains, the southwest regions （excluding the Chengdu Plain）, Loess Plateau and south- east coastal regions, had a lower RRA, being less than 20%. Moreover, RRA varied with topographic slope and altitude. In the areas of low altitude （~〈250 m）, middle altitude （250-100 m） and high altitude （1000-3500 m）, there were 443 million, 215 million and 64 million mu of cropland respectively and their regional mean RRAs were 27.5%, 12.6% and 7.2% respectively. In the areas of flat slope, gentle slope, medium slope and steep slope, there were 116 million, 456 million, 144 million and 2 million mu of cropland respectively and their regional mean RRAs were 34.6%, 20.7%, 8.5% and 2.3% respectively.
