Based on a 2-D hydrodynamic model, a vertically integrated eutrophication model was developed. The physical sub-model can be used for calculation of water density at different depths, and the water quality sub-model w...Based on a 2-D hydrodynamic model, a vertically integrated eutrophication model was developed. The physical sub-model can be used for calculation of water density at different depths, and the water quality sub-model was used for calculation of algal growth. The cohesive and non-cohesive sediments were simulated separately with different methods. The light extinction coefficient used in the underwater light regime sub-model was linearly related to the sum of sediment and phytoplankton biomass. Some components less important to the model were simplified to improve practicability and calculation efficiency. Using field data from Fuchunjiang Reservoir, we calculated the sensitivity of ecological parameters included in this model and validated the model. The results of sensitivity analysis showed that the parameters strongly influenced the phytoplankton biomass, including phytoplankton maximum growth rate, respiration rate, non-predatory mortality rate, settling rate, zooplankton maximum filtration rate, specific extinction coefficient for suspended solids and sediment oxygen demand rate. The model was calibrated by adjusting these parameters. Total chlorophyll α （chl-α） concentrations at different layers in the water column were reproduced very well by the model simulations. The simulated chl-α values were positively correlated to the measured values with Pearson correlation coefficient of 0.92. The mean difference between measured and simulated chl-α concentrations was 12% of the measured chl-α concentration. Measured and simulated DO concentrations were also positively correlated （r = 0.74） and the mean difference was 4% of measured DO concentrations. The successful validation of model indicated that it would be very useful in water quality management and algal bloom prediction in Fuchunjiang Reservoir and a good tool for water quality regulation of other fiver-style reservoirs.展开更多
针对服装风格人工分类受主观性、地域等因素影响而造成的分类错误问题,研究了一种基于人工智能的服装风格图像分类方法。首先,在FashionStyle14数据集基础上筛除重复或无效图像,构建服装风格图像数据集;然后,采用迁移学习方法,对Efficie...针对服装风格人工分类受主观性、地域等因素影响而造成的分类错误问题,研究了一种基于人工智能的服装风格图像分类方法。首先,在FashionStyle14数据集基础上筛除重复或无效图像,构建服装风格图像数据集;然后,采用迁移学习方法,对EfficientNet V2、RegNet Y 16GF和ViT Large 16等模型进行微调训练,生成新模型,实现基于单个深度学习的服装风格图像分类;最后,为进一步提高图像分类的准确性、可靠性和鲁棒性,分别采用基于投票、加权平均和堆叠的集成学习方法对上述单个模型进行组合预测。迁移学习实验结果表明,基于ViT Large 16的深度学习模型在测试集上表现最佳,平均准确率为77.024%;集成学习方法实验结果显示,基于投票的集成学习方法在相同测试集上平均准确率可达78.833%。研究结果为解决服装风格分类问题提供了新的思路。展开更多
基金supported by the National Natural Sci-ence Foundation of China (No. 40730529, 40501078)the Chinese Academy of Sciences (No. KZCX2-YW-419)the Department of Science and Technology of Zhejiang Province (No. 2005C13001)
文摘Based on a 2-D hydrodynamic model, a vertically integrated eutrophication model was developed. The physical sub-model can be used for calculation of water density at different depths, and the water quality sub-model was used for calculation of algal growth. The cohesive and non-cohesive sediments were simulated separately with different methods. The light extinction coefficient used in the underwater light regime sub-model was linearly related to the sum of sediment and phytoplankton biomass. Some components less important to the model were simplified to improve practicability and calculation efficiency. Using field data from Fuchunjiang Reservoir, we calculated the sensitivity of ecological parameters included in this model and validated the model. The results of sensitivity analysis showed that the parameters strongly influenced the phytoplankton biomass, including phytoplankton maximum growth rate, respiration rate, non-predatory mortality rate, settling rate, zooplankton maximum filtration rate, specific extinction coefficient for suspended solids and sediment oxygen demand rate. The model was calibrated by adjusting these parameters. Total chlorophyll α （chl-α） concentrations at different layers in the water column were reproduced very well by the model simulations. The simulated chl-α values were positively correlated to the measured values with Pearson correlation coefficient of 0.92. The mean difference between measured and simulated chl-α concentrations was 12% of the measured chl-α concentration. Measured and simulated DO concentrations were also positively correlated （r = 0.74） and the mean difference was 4% of measured DO concentrations. The successful validation of model indicated that it would be very useful in water quality management and algal bloom prediction in Fuchunjiang Reservoir and a good tool for water quality regulation of other fiver-style reservoirs.
文摘针对服装风格人工分类受主观性、地域等因素影响而造成的分类错误问题,研究了一种基于人工智能的服装风格图像分类方法。首先,在FashionStyle14数据集基础上筛除重复或无效图像,构建服装风格图像数据集;然后,采用迁移学习方法,对EfficientNet V2、RegNet Y 16GF和ViT Large 16等模型进行微调训练,生成新模型,实现基于单个深度学习的服装风格图像分类;最后,为进一步提高图像分类的准确性、可靠性和鲁棒性,分别采用基于投票、加权平均和堆叠的集成学习方法对上述单个模型进行组合预测。迁移学习实验结果表明,基于ViT Large 16的深度学习模型在测试集上表现最佳,平均准确率为77.024%;集成学习方法实验结果显示,基于投票的集成学习方法在相同测试集上平均准确率可达78.833%。研究结果为解决服装风格分类问题提供了新的思路。
