Matching soil grid unit resolutions with polygon unit map scales is important to minimize the uncertainty of regional soil organic carbon(SOC) pool simulation due to their strong influences on the modeling.A series of...Matching soil grid unit resolutions with polygon unit map scales is important to minimize the uncertainty of regional soil organic carbon(SOC) pool simulation due to their strong influences on the modeling.A series of soil grid units at varying cell sizes was derived from soil polygon units at six map scales,namely,1:50 000(C5),1:200 000(D2),1:500 000(P5),1:1 000 000(N1),1:4 000 000(N4) and 1:14 000 000(N14),in the Taihu Region of China.Both soil unit formats were used for regional SOC pool simulation with a De Nitrification-DeC omposition(DNDC) process-based model,which spans the time period from 1982 to 2000 at the six map scales.Four indices,namely,soil type number(STN),area(AREA),average SOC density(ASOCD) and total SOC stocks(SOCS) of surface paddy soils that were simulated by the DNDC,were distinguished from all these soil polygon and grid units.Subjecting to the four index values(IV) from the parent polygon units,the variations in an index value(VIV,%) from the grid units were used to assess its dataset accuracy and redundancy,which reflects the uncertainty in the simulation of SOC pools.Optimal soil grid unit resolutions were generated and suggested for the DNDC simulation of regional SOC pools,matching their respective soil polygon unit map scales.With these optimal raster resolutions,the soil grid units datasets can have the same accuracy as their parent polygon units datasets without any redundancy,when VIV < 1% was assumed to be a criterion for all four indices.A quadratic curve regression model,namely,y = – 0.80 × 10^(–6)x^2 + 0.0228 x + 0.0211(R^2 = 0.9994,P < 0.05),and a power function model R? = 10.394?^(0.2153)(R^2 = 0.9759,P < 0.05) were revealed,which describe the relationship between the optimal soil grid unit resolution(y,km) and soil polygon unit map scale(1:10 000x),the ratio(R?,%) of the optimal soil grid size to average polygon patch size(?,km^2) and the ?,with the highest R^2 among different mathematical regressions,respectively.This knowledge may facilitate the grid partitioning of regions during the investigation and simulation of SOC pool dynamics at a certain map scale,and be referenced to other landscape polygon patches' mesh partition.展开更多
为了给驾驶员提供实时准确的行人信息、减少交通事故的发生,提出一种检测增强型YOLOv3-tiny(detection of enhanced YOLOv3-tiny,DOEYT)行人检测算法.创建鲁棒的特征提取网络,首先使用非对称最大池化进行下采样,防止随着感受野增大行人...为了给驾驶员提供实时准确的行人信息、减少交通事故的发生,提出一种检测增强型YOLOv3-tiny(detection of enhanced YOLOv3-tiny,DOEYT)行人检测算法.创建鲁棒的特征提取网络,首先使用非对称最大池化进行下采样,防止随着感受野增大行人横向特征的丢失;其次使用Hardswish作为卷积层的激活函数优化网络性能;最后使用GC(globe context)自注意力机制获得全文特征信息.在分类回归网络部分,采用三尺度检测策略,提升小尺度行人目标的检测精度;使用k-means++算法重新生成数据集锚框,提高网络收敛速度.构建行人检测数据集并分为训练集和测试集,对DOEYT算法的性能进行试验验证.结果表明,非对称最大池化、Hardswish函数、GC自注意力机制分别使平均准确率AP提高14.4%、7.9%、10.8%;DOEYT算法在测试集上检测的平均准确率高达91.2%,检测速度为103帧/s,可见该算法可快速准确地检测行人,降低交通事故发生的风险.展开更多
基金Under the auspices of Special Project of National Key Research and Development Program(No.2016YFD0200301)National Natural Science Foundation of China(No.41571206)Special Project of National Science and Technology Basic Work(No.2015FY110700-S2)
文摘Matching soil grid unit resolutions with polygon unit map scales is important to minimize the uncertainty of regional soil organic carbon(SOC) pool simulation due to their strong influences on the modeling.A series of soil grid units at varying cell sizes was derived from soil polygon units at six map scales,namely,1:50 000(C5),1:200 000(D2),1:500 000(P5),1:1 000 000(N1),1:4 000 000(N4) and 1:14 000 000(N14),in the Taihu Region of China.Both soil unit formats were used for regional SOC pool simulation with a De Nitrification-DeC omposition(DNDC) process-based model,which spans the time period from 1982 to 2000 at the six map scales.Four indices,namely,soil type number(STN),area(AREA),average SOC density(ASOCD) and total SOC stocks(SOCS) of surface paddy soils that were simulated by the DNDC,were distinguished from all these soil polygon and grid units.Subjecting to the four index values(IV) from the parent polygon units,the variations in an index value(VIV,%) from the grid units were used to assess its dataset accuracy and redundancy,which reflects the uncertainty in the simulation of SOC pools.Optimal soil grid unit resolutions were generated and suggested for the DNDC simulation of regional SOC pools,matching their respective soil polygon unit map scales.With these optimal raster resolutions,the soil grid units datasets can have the same accuracy as their parent polygon units datasets without any redundancy,when VIV < 1% was assumed to be a criterion for all four indices.A quadratic curve regression model,namely,y = – 0.80 × 10^(–6)x^2 + 0.0228 x + 0.0211(R^2 = 0.9994,P < 0.05),and a power function model R? = 10.394?^(0.2153)(R^2 = 0.9759,P < 0.05) were revealed,which describe the relationship between the optimal soil grid unit resolution(y,km) and soil polygon unit map scale(1:10 000x),the ratio(R?,%) of the optimal soil grid size to average polygon patch size(?,km^2) and the ?,with the highest R^2 among different mathematical regressions,respectively.This knowledge may facilitate the grid partitioning of regions during the investigation and simulation of SOC pool dynamics at a certain map scale,and be referenced to other landscape polygon patches' mesh partition.
文摘为了给驾驶员提供实时准确的行人信息、减少交通事故的发生,提出一种检测增强型YOLOv3-tiny(detection of enhanced YOLOv3-tiny,DOEYT)行人检测算法.创建鲁棒的特征提取网络,首先使用非对称最大池化进行下采样,防止随着感受野增大行人横向特征的丢失;其次使用Hardswish作为卷积层的激活函数优化网络性能;最后使用GC(globe context)自注意力机制获得全文特征信息.在分类回归网络部分,采用三尺度检测策略,提升小尺度行人目标的检测精度;使用k-means++算法重新生成数据集锚框,提高网络收敛速度.构建行人检测数据集并分为训练集和测试集,对DOEYT算法的性能进行试验验证.结果表明,非对称最大池化、Hardswish函数、GC自注意力机制分别使平均准确率AP提高14.4%、7.9%、10.8%;DOEYT算法在测试集上检测的平均准确率高达91.2%,检测速度为103帧/s,可见该算法可快速准确地检测行人,降低交通事故发生的风险.
