Despite the improvement in cultivar characters and management practices, large gaps between the attainable and potential yields still exist in winter wheat of China. Quantifying the crop potential yield is essential for estimating the food production capacity and improving agricultural policies to ensure food security. Gradually descending models and geographic infor- mation system （GIS） technology were employed to characterize the spatial variability of potential yields and yield gaps in winter wheat across the main production region of China. The results showed that during 2000-2010, the average potential yield limited by thermal resource （YGT） was 23.2 Mg ha-1, with larger value in the northern area relative to the southern area. The potential yield limited by the water supply （YGw） generally decreased from north to south, with an average value of 1.9 Mg ha-1 across the entire study region. The highest YGw in the north sub-region （NS） implied that the irrigation and drainage conditions in this sub-region must be improved. The averaged yield loss of winter wheat from nutrient deficiency （YGH） varied between 2.1 and 3.1 Mg ha-1 in the study area, which was greater than the yield loss caused by water limitation. The potential decrease in yield from photo-thermal-water-nutrient-limited production to actual yield （YGo） was over 6.0 Mg ha-1, ranging from 4.9 to 8.3 Mg ha^-1 across the entire study region, and it was more obvious in the southern area than in the northern area. These findings suggest that across the main winter wheat production region, the highest yield gap was induced by thermal resources, followed by other factors, such as the level of farming technology, social policy and economic feasibility. Furthermore, there are opportunities to narrow the yield gaps by making full use of climatic resources and developing a reasonable production plan for winter wheat crops. Thus, meeting the challenges of food security and sustainability in the coming decades is possible but will require
针对多标记迁移学习中源领域与目标领域的特征分布差异会导致源领域数据无法被目标领域利用的问题,提出了一种基于最大均值差异的多标记迁移学习算法(Multi-Label Transfer Learning via Maximum mean discrepancy,M-MLTL),算法通过分解关系矩阵构造共享子空间,并采用最大均值差异(maximum mean discrepancy)作为评价指标,最小化子空间特征的分布差异,从而使源领域与目标领域的特征分布尽可能相似.多标记图像分类实验的结果表明,新算法比同类算法有更高的精度和计算效率.
