基于迁移学习的棉花叶部病虫害图像识别

针对传统图像识别方法准确率低、手工提取特征等问题,该研究以棉花叶部病虫害图像为研究对象,利用迁移学习算法并辅以数据增强技术,实现棉花叶部病虫害图像准确分类。首先改进AlexNet模型,利用PlantVillage大数据集训练取得预训练模型,在预训练模型上使用棉花病虫害数据微调参数,得到平均测试准确率为93.50%;然后使用数据增强技术扩充原始数据集,在预训练模型上再训练,得到最终平均测试准确率为97.16%。相同试验条件下,该研究方法较支持向量机(Support Vector Machine,SVM)和BP(Back Propagation,BP)神经网络以及深度卷积模型(VGG-19和GoogLeNet Inception v2)分类效果更好。试验结果表明,通过迁移学习能把从源领域(PlantVillage数据集)学习到的知识迁移到目标领域(棉花病虫害数据集),数据增强技术能有效缓解过拟合。该研究为农作物病虫害识别技术的发展提供了参考。     

黄河未来输沙量态势及其适用性对策

[目的] 探索黄河输沙预测的新思路，预估黄河未来输沙态势与输沙量水平，为黄河流域生态治理规划提供参考。[方法] 结合黄河流域水土保持生态修复现状，采用单累积曲线法、滑动平均及频率分析方法，分析1950—2019年黄河主要来沙区间的实测输沙量变化特征及其未来态势。[结果] 1950—2019年黄河输沙量呈现阶梯式减少。1950—2019年黄河中游各站累积实测输沙量随时间的变化可用“左半抛物线”表征。黄河输沙量自1997年以来已进入相对稳定态势，目前已达企稳状态；黄河潼关站未来年输沙量在90%频率下为1.00×10⁸ t左右，在10%频率下为5.00×10⁸ t左右，未来多年平均输沙量为1.40×10⁸ t。[结论] 为了维持黄河输沙量低稳状态，提升水土保持措施质量与标准，补齐“后水土保持”短板，构建完善的水沙关系调控体系，维持黄河下游河道冲淤平衡，是黄河流域生态保护与高质量发展的保障。     

川渝地区地方和育成大豆品种SSR标记多样性分析?

利用基本均匀分布于大豆20条染色体的135对SSR标记，对232份包括6个地方品种地域亚群和1个育成品种亚群进行全基因组扫描。结果表明：所有的标记都有多态性，所有检测到的位点都是纯合基因型，说明所选用品种高度纯合，每个标记存在2～4个等位变异，平均2.66个。亚群多态信息含量变异范围0.2751-0.3165，整个群体为0.3208；亚群内Nei遗传距离变异范围0.325 8～0.359 4，整个群体为0.3711，说明川渝地区大豆遗传变异较小。亚群间的遗传一致度（GI ≥ 0.8862）较高，亚群间遗传距离（GD ≤ 0.1208）较小，地方品种亚群间遗传差异更小，育成品种亚群与自然地域亚群的遗传差异相对较大。亚群间基因分化系数（Fst）平均为 0.0722，基因流（Nm）平均为 3.214，说明不同亚群之间存在一定的基因交流。主坐标分析表明第一、二和三主成分分别解释总变异的4.97%、3.54%和3.33%。来自同一区域的品种资源基本聚集在同一亚群，聚类分析同样表明同一自然地域亚群品种资源虽不能完全聚集到同一个遗传类群中，但具有一定的聚集效应，说明川渝大豆品种资源遗传变异与地理位置有一定的关系。分子方差分析表明亚群内变异占总变异的97%，亚群间变异仅占总变异的3%。Mantel收敛分析表明地方品种自然地域亚群的遗传距离与所处的地理位置距离（纬度和海拔）呈显著的正相关关系（R2=0.723）。川渝地区大豆种质资源群体遗传丰富度不高，当前的育成品种未蕴含本地区所有遗传变异。     

Monitoring desertification processes in Mongolian Plateau using MODIS tasseled cap transformation and TGSI time series

Most remote sensing studies assess the desertification using vegetation monitoring method. But it has the insufficient precision of vegetation monitoring for the limited vegetation cover of the desertification region. Therefore, it offers an alternative approach for the desertification research to assess sand dune and sandy land change using remote sensing in the desertification region. In this study, the indices derived from the well-known tasseled cap transformation(TCT), tasseled cap angle(TCA),disturbance index(DI), process indicator(PI), and topsoil grain size index(TGSI) were integrated to monitor and assess the desertification at the thirteen study sites including sand dunes and sandy lands distributed in the Mongolian Plateau(MP) from 2000 to 2015. A decision tree was used to classify the desertification on a regional scale. The average overall accuracy of 2000, 2005, 2010 and 2015 desertification classification was higher than 90%. Results from this study indicated that integration of the advantages of TCA, DI and TGSI could better assess the desertification. During the last 16 years, Badain Jaran Desert, Tengger Desert, and Ulan Buh Desert showed a relative stabilization. Otindag Sandy Land and the deserts of Khar Nuur, Ereen Nuur, Tsagan Nuur, Khongoryn Els, Hobq, and Mu Us showed a slow increasing of desertification, whereas Bayan Gobi, Horqin and Hulun Buir sandy lands showed a slow decreasing of desertification. Compared with the other 11 sites, the fine sand dunes occupied the majority of the Tengger Desert, and the coarse sandy land occupied the majority of the Horqin Sandy Land. Our findings on a three or four years' periodical fluctuated changes in the desertification may possibly reflect changing precipitation and soil moisture in the MP. Further work to link the TCA, DI,TGSI, and PI values with the desertification characteristics is recommended to set the thresholds and improve the assessment accuracy with field investigation.     

西方语境的中国古典园林审美认知历史溯源

中国古典园林审美不仅仅在中国文化体系中进行讨论，随着全球化进程，西方语境学者也逐渐涉及相关的内容。尝试寻找一种跨文化的“他者”视角，梳理近3个世纪以来西方语境的中国古典园林审美认知经历的3个主要阶段。以18世纪为开端，经过19—20世纪借助西译艺术著作和海外园林建设实现对审美理解从印象到真实的转变，到进入21世纪后诞生的跨学科的诸多新观，这个过程中展现出以中国传统园林审美为代表的打破自身线性历史的闭关自守，回归全球语境的未来中国古典园林研究趋势。     

Effects of different tillage and straw retention practices on soil aggregates and carbon and nitrogen sequestration in soils of the northwestern China

Soil tillage and straw retention in dryland areas may affect the soil aggregates and the distribution of total organic carbon. The aims of this study were to establish how different tillage and straw retention practices affect the soil aggregates and soil organic carbon (SOC) and total nitrogen (TN) contents in the aggregate fractions based on a long-term (approximately 15 years) field experimentin the semi-arid western Loess Plateau, northwestern China. The experiment included four soil treatments, i.e., conventional tillage with straw removed (T), conventional tillage with straw incorporated (TS), no tillage with straw removed (NT) and no tillage with straw retention (NTS), which were arranged in a complete randomized block design. The wet-sieving method was used to separate four size fractions of aggregates, namely, large macroaggregates (LA, >2000 μm), small macroaggregates (SA, 250-2000 μm), microaggregates (MA, 53-250 μm), and silt and clay (SC, <53 μm). Compared to the conventional tillage practices (including T and TS treatments), the percentages of the macroaggregate fractions (LA and SA) under the conservation tillage practices (including NT and NTS treatments) were increased by 41.2%-56.6%, with the NTS treatment having the greatest effect. For soil layers of 0-5, 5-10 and 10-30 cm, values of the mean weight diameter (MWD) under the TS and NTS treatments were 10.68%, 13.83% and 17.65%, respectively. They were 18.45%, 19.15% and 14.12% higher than those under the T treatment, respectively. The maximum contents of the aggregate-associated SOC and TN were detected in the SA fraction, with the greatest effect being observed for the NTS treatment. The SOC and TN contents were significantly higher under the NTS and TS treatments than under the T treatment. Also, the increases in SOC and TN levels were much higher in the straw-retention plots than in the straw-removed plots. The macroaggregates (including LA and SA fractions) were the major pools for SOC and TN, regardless of tillage practices, storing 3.25-6.81 g C/kg soil and 0.34-0.62 g N/kg soil. Based on the above results, we recommend the NTS treatment as the best option to boost soil aggregates and to reinforce carbon and nitrogen sequestration in soils in the semi-arid western Loess Plateau of northwestern China.     

331份四川小麦品种(系)在甘肃陇南抗条锈性表现及利用价值

2014年-2017年度先后对331份四川省小麦生产品种(系)在温室进行苗期人工接种条锈菌混合菌鉴定和甘谷试验站大田成株期分别接种CYR32、CYR33、CYR34、G22-14、中4-1和混合菌鉴定,同时在甘谷试验站和汪川良种场两地进行自然诱发鉴定。结果发现:在人工接种条件下,苗期、成株期表现抗病的分别有‘XK201465’等36份和‘XK20132’等72份材料,分别占10.88%和21.75%;有‘XK62483’等11份材料全生育期表现抗病,占3.32%;两地三年自然诱发鉴定发现,仅有‘XK20132’等5份材料在两地均表现抗病,有‘川农17’等30份材料具有慢条锈性。对供鉴材料在甘肃陇南的利用前景进行了分析。     

3种遥感产品检测黄土高原森林变化及差异比较

基于Landsat Vegetation Continuous Fields（LVCF）、MODIS Vegetation Continuous Fields（MVCF）和MODIS Land Cover（MODLC）等遥感产品和森林清查数据（NFI）检测2000-2015年黄土高原森林分布及变化。结果表明:1）2000-2015年黄土高原森林覆盖度升高，基于LVCF和MVCF的全区森林覆盖度均值分别由2000年的7.8%和9.6%增加到2015年的9.7%和13.2%。2）遥感产品估算的森林分布空间格局和面积差异较大，基于LVCF、MVCF和MODLC估算的黄土高原森林面积在2000年分别为726.6×10⁴、604.7×10⁴ hm²和325.1×10⁴ hm²，2015年分别为926.2×10⁴、998.3×10⁴ hm²和400.1×10⁴ hm²。3）遥感产品估算的省级尺度上森林面积与NFI的不一致性因省份和年份而异，LVCF与NFI一致性整体略优于MVCF，MODLC与NFI差异最大且估算的森林面积远低于LVCF、MVCF和NFI。4）遥感产品检测的陕西和山西2省的森林面积增加值与NFI一致性高于宁夏；基于遥感产品估算的宁夏森林面积及增加幅度均远低于NFI。     

青藏高原东部强降温过程阈值的探讨及其气候特征

为解决青藏高原东部强降温的气象预报和服务问题,利用青藏高原东北部68个气象站1961-2017年逐日最低气温数据,辅以趋势分析及相关分析等方法,本文探讨了近57年青藏高原东北部强降温的阈值及气候特征,并利用NECP/NCAR(2.5°×2.5°)海温场、高度场建立了预报诊断模型。研究发现 (1) 应用改进的强降温指标统计寒潮发生的特征更符合高原实际情况。(2)受地形地貌及山脉阻挡的影响,青藏高原东部强降温次数的地域分布存在明显差异,海拔较高的山区和东部地区强降温次数明显多于盆地。南部山区24h寒潮发生的次数多于48h、72h。(3)近57年区域寒潮频次总体呈下降趋势,南部和北部的降速分别为-0.5次/10a、-0.4次/10a,并通过显著性检验；西北部盆地区降速分别为-0.1次/10a,未通过显著性检验。(4)高冷空气的移动路径是强降温预测的关键信号。研究结果为青藏高原相关部门开展农牧业资源的和旅游资源的相关决策服务提供科学依据。     

播期对川中丘陵区油菜倒伏性状的影响

油菜机械化生产中,植株倒伏是影响机械化收获的重要因素。为探讨不同播种期对川中丘陵区油菜倒伏性状的影响,以川油36和蓉油18为供试材料,分别设置9月22日、9月29日、10月6日、10月13日和10月20日5个播期处理,分析不同播期条件下油菜生育期气象因子、倒伏性状指标的变化情况。结果表明,随播种期推迟,油菜生育期内的降水量、积温、日均最高气温、日均最低气温和日照时长均逐渐下降,生育期逐渐缩短,成熟期植株株高、分枝部位、根颈粗、茎秆粗降低,倒伏系数增加。油菜抗倒伏能力在9月22-29日播期处理之间最强。因此,川中丘陵区油菜的适宜播期为9月下旬。