基于卷积神经网络马铃薯叶片病害识别和病斑检测*

针对自然环境下马铃薯叶片病害识别率低和晚疫病斑定位难的问题,基于大田环境中采集的马铃薯叶片图像,首先对马铃薯叶片病害进行识别,对比AlexNet、VGG16、InceptionV3、ResNet50、MobileNet五种神经网络模型,结果表明InceptionV3模型的识别效果准确率最高,可达98.00%。其次对马铃薯叶片的晚疫病斑进行检测,提出一种改进型的CenterNet-SPP模型,该模型通过特征提取网络获取对象的中心点,再通过中心点回归获得中心点偏移量、目标大小等图像信息,训练后的模型在验证集下的mAP可达90.03%,以F₁为评价值分析对比其它目标检测模型,CenterNet-SPP模型的效果最好,准确率为94.93%,召回率为90.34%,F₁值为92.58%,平均检测一张图像耗时0.10 s。为自然环境下马铃薯叶片病害识别和检测提供较为全面的深度学习算法和模型研究基础。     

紫云英绿肥盛花期埋切翻压组合作业机设计与试验

为实现紫云英绿肥规模化种植区域的绿肥高效、低耗机械化翻压,基于前端埋切、后端翻压的结构配置思路,设计了一种无动力输入的紫云英绿肥盛花期埋切翻压组合作业机。对盛花期紫云英生物学特征及压切特性开展研究,进而设计了辊筒组件、前悬挂连接组件、栅条式翻转犁等主要工作部件;通过理论计算和力学分析,得到实现紫云英茎秆顺畅埋切的压切辊半径为305 mm、压切刀滑切角范围为34. 3°～55. 7°;采用水平直元线法设计了栅条式犁体曲面,推导得到导曲线抛物线方程,并绘制了犁体直元线角变化规律图。紫云英绿肥盛花期田间翻压试验结果表明:在机具作业速度为5. 8～6. 0 km/h条件下,茎秆切断长度合格率92%,耕宽稳定性变异系数3. 7%,耕深稳定性变异系数5. 6%,土垡破碎率85. 6%,植被翻压覆盖率98. 3%,试验指标均符合国家和行业标准要求。设计的埋切翻压组合作业机工作效率可达0. 67～1. 20 hm2/h,在翻压质量和组配方式上明显优于目前我国常用的畜力犁、铧式犁、旋耕机等翻压方式,具有很好的推广应用价值。     

Deep neural network algorithm for estimating maize biomass based on simulated Sentinel 2A vegetation indices and leaf area index

Accurate estimation of biomass is necessary for evaluating crop growth and predicting crop yield.Biomass is also a key trait in increasing grain yield by crop breeding.The aims of this study were(i)to identify the best vegetation indices for estimating maize biomass,(ii)to investigate the relationship between biomass and leaf area index(LAI)at several growth stages,and(iii)to evaluate a biomass model using measured vegetation indices or simulated vegetation indices of Sentinel 2A and LAI using a deep neural network(DNN)algorithm.The results showed that biomass was associated with all vegetation indices.The three-band water index(TBWI)was the best vegetation index for estimating biomass and the corresponding R2,RMSE,and RRMSE were 0.76,2.84 t ha−1,and 38.22%respectively.LAI was highly correlated with biomass(R2=0.89,RMSE=2.27 t ha−1,and RRMSE=30.55%).Estimated biomass based on 15 hyperspectral vegetation indices was in a high agreement with measured biomass using the DNN algorithm(R2=0.83,RMSE=1.96 t ha−1,and RRMSE=26.43%).Biomass estimation accuracy was further increased when LAI was combined with the 15 vegetation indices(R2=0.91,RMSE=1.49 t ha−1,and RRMSE=20.05%).Relationships between the hyperspectral vegetation indices and biomass differed from relationships between simulated Sentinel 2A vegetation indices and biomass.Biomass estimation from the hyperspectral vegetation indices was more accurate than that from the simulated Sentinel 2A vegetation indices(R2=0.87,RMSE=1.84 t ha−1,and RRMSE=24.76%).The DNN algorithm was effective in improving the estimation accuracy of biomass.It provides a guideline for estimating biomass of maize using remote sensing technology and the DNN algorithm in this region.     

Research of dissolved oxygen prediction in recirculating aquaculture systems based on deep belief network

Recirculating aquaculture has received more and more attention because of its high efficiency of treatment and recycling of aquaculture wastewater. The content of dissolved oxygen is an important indicator of control in recirculating aquaculture, its content and dynamic changes have great impact on the healthy growth of fish. However, changes of dissolved oxygen content are affected by many factors, and there is an obvious time lag between control regulation and effects of dissolved oxygen. To ensure the aquaculture production safety, it is necessary to predict the dissolved oxygen content in advance. The prediction model based on deep belief network has been proposed in this paper to realize the dissolved oxygen content prediction. A variational mode decomposition (VMD) data processing method has been adopted to evaluate the original data space, it takes the data which has been decomposed by the VMD as the input of deep belief network (DBN) to realize the prediction. The VMD method can effectively separate and denoise the raw data, highlight the relations among data features, and effectively improve the quality of the neural network input. The proposed model can quickly and accurately predict the dissolved oxygen content in time series, and the prediction performance meets the needs of actual production. When compared with bagging, AdaBoost, decision tree and convolutional neural network, the VMD-DBN model produces higher prediction accuracy and stability.     

一种新型卷积神经网络植物叶片识别方法

为提高植物叶片识别的准确率及减少计算代价,在Pytorch框架下提出一种融合了深度卷积生成式对抗网络(DCGAN)和迁移学习(TL)的新型卷积神经网络叶片识别方法。首先,对植物叶片图像进行预处理,通过DCGAN对样本数据库扩充;其次,利用迁移学习将Inception v3模型应用于图像数据处理上,以提高植物叶片识别的准确率;最后,通过对比实验对该方法的有效性进行验证。结果表明:该方法可以获得96.57%的植物叶片识别精度,同时参数训练的迭代次数由4000次缩短到560次。     

Early life history connectivity of Antarctic silverfish (Pleuragramma antarctica) in the Ross Sea

A recent population hypothesis for Antarctic silverfish (Pleuragramma antarctica), a critical forage species, argued that interactions between life history and circulation associated with glacial trough systems drive circumpolar distributions over the continental shelf. In the Ross Sea, aggregations of eggs and larvae occur under fast ice in Terra Nova Bay, and the hypothesis predicted that dispersing larvae encounter outflow along the western side of Drygalski Trough. The outflow advects larvae towards the shelf‐break, and mixing with trough inflow facilitates return toward the inner shelf. To examine the hypothesis, we compared samples of P. antarctica collected near Coulman Island in the outflow, along Crary Bank in the inflow, and a third set taken over the rest of the Ross Sea. We ruled out misidentification using an innovative genetic validation. Silverfish larvae comprised 99.5% of the catch, and the highest population densities were found in Drygalski Trough. The results provided no evidence to reject the population hypothesis. Abundance indices, back‐calculated hatching dates, length distributions and growth were congruent with a unified early life history in the western Ross Sea, constrained by cryopelagic early stages in Terra Nova Bay. By contrast, a sample in the Bay of Whales revealed much smaller larvae, suggesting either a geographically separate population in the eastern Ross Sea, or westward connectivity with larvae spawned nearby by fish sourced from troughs upstream in the Amundsen Sea. These results illustrate how hypotheses that integrate population structure with life history can provide precise spatial predictions for subsequent testing.     

4种水生植物深度净化村镇生活污水厂尾水效果研究

通过设置动态模拟试验,持续进水、出水条件下分析比较了漂浮植物凤眼莲和水浮莲、沉水植物轮叶黑藻和挺水植物黄菖蒲对村镇生活污水厂(一级A标准)尾水深度净化效果,筛选出具有去污效果优势的水生植物,为优化水生植物生态修复工程技术在尾水深度净化中的应用提供依据。结果表明:经水生植物深度净化后,尾水水质得到明显改善,漂浮植物凤眼莲和水浮莲对尾水氮、磷的净化效果优于挺水植物黄菖蒲和沉水植物轮叶黑藻。试验周期内,污水厂尾水总氮、总磷和高锰酸盐指数(CODMn)平均浓度为12.22 mg?L-1、0.38mg?L-1和3.88 mg?L-1,凤眼莲、水浮莲、轮叶黑藻、黄菖蒲和对照各系统的总氮平均去除率分别为46.25%、45.74%、43.41%、38.39%和29.22%,总磷去除率分别为36.84%、34.21%、31.58%、28.95%和26.32%,CODMn去除率分别为42.27%、30.93%、32.47%、32.47%和37.89%。凤眼莲、水浮莲、黄菖蒲和轮叶黑藻生物量净增长率分别为550.5%、418.8%、210.6%和80.3%,凤眼莲生物量净增率最大。各处理系统内凤眼莲、水浮莲、黄菖蒲和轮叶黑藻对尾水氮富集量分别为7.36 g、2.33 g、5.12 g和4.46 g,对磷的富集量分别为0.60 g、0.19 g、0.33 g和0.78 g,凤眼莲富集氮能力优于另外3种水生植物,轮叶黑藻磷富集量高于另外3种水生植物。凤眼莲、水浮莲、黄菖蒲和轮叶黑藻植株吸收作用对尾水总氮去除的表观贡献率分别为15.29%、4.90%、11.17%和11.34%,对尾水总磷去除的表观贡献率分别为50.34%、17.17%、35.24%和76.34%。因此,可利用漂浮植物凤眼莲和沉水植物轮叶黑藻立体复合种养的方式深度净化生活污水厂尾水。     

保水剂与秸秆深施对砂质潮土水肥状况及小麦产量的影响

保水剂与秸秆深施对砂质潮土培肥效应的研究,为中低产田砂质潮土的改良提供科学合理的依据。于典型的砂质潮土区设置了五种不同的田间处理方式:秸秆翻埋还田(TS)、秸秆深施(DS)、秸秆与保水剂深施(DS+W)、保水剂深施(DW)和不处理对照(CK)。比较分析了小麦生育期不同土层中水分、有机质、速效氮以及小麦产量。研究结果表明:(1)与CK相比,从孕穗期到灌浆期,秸秆与保水剂的施用均使0~80 cm土层的含水量有所提高,且DS+W与DW均显著提高了60~80 cm土层水分,其中在孕穗期两处理水分含量分别是CK的3倍和2倍,灌浆期为CK的2倍和1.5倍。(2)与CK相比,其余各处理均显著提高了小麦产量(P<0.05),TS、DS和DS+W分别增产了30.7%、27.3%和48.9%。(3)与CK相比,TS、DS、DS+W和DW的水分利用效率显著增加,分别增加了0.63、0.57 0.93和0.29 kg m~(-3)。(4)与CK相比,20~80 cm土层中,深埋处理DS、DS+W、DW均提高了土壤中有机质含量,DS+W处理下土壤有机质含量与其他处理之间差异显著(P<0.05)。(5)与CK相比,TS和DS处理能够有效的提高各土层中硝态氮含量,保水剂的施用在40~80 cm土层中对土壤硝态氮未见显著影响。(6)各处理间单茎小麦各器官干物质积累量大小为:DS+W>DS>TS>DW>CK,且从不同器官的干物质积累量的比例来看,TS、DS、DS+W处理下的籽粒所占比重较CK的籽粒所占比重有所增加,茎+叶所占比例相应减少,植株干物质的积累逐渐向籽粒偏斜。(7)从小麦产量与土壤水分和硝态氮含量的相关分析结果来看,土壤水分主要在小麦灌浆期影响小麦产量,且是砂质潮土中影响小麦产量的最主要因素。综上所述,保水剂与秸秆的施用均能不同程度的改善砂质潮土的理化性状,提高农作物产量,以保水剂与秸秆的配合深施对砂质潮土的改良效果最佳。     

Soil Nitrogen Distribution and Plant Nitrogen Utilization in Direct-Seeded Rice in Response to Deep Placement of Basal Fertilizer-Nitrogen

Deep placement of controlled-release fertilizer increases nitrogen(N) use efficiency in rice planting but is expensive. Few studies on direct-seeded rice have examined the effects of deep placement of conventional fertilizer. With prilled urea serving as N fertilizer, a two-year field experiment with two N rates(120 and 195 kg/hm~2) and four basal N application treatments(B50, all fertilizer was broadcast with 50% as basal N; D50, D70 and D100 corresponded to 50%, 70% and 100% of N deeply placed as basal N, respectively) were conducted in direct-seeded rice in 2013 and 2014. Soil N distribution and plant N uptake were analyzed. The results showed that deep placement of basal N significantly increased total N concentrations in soil. Significantly greater soil N concentrations were observed in D100 compared with B50 at 0, 6 and 12 cm(lateral distance) from the fertilizer application point both at mid-tillering and heading stages. D100 presented the highest values of dry matter and N accumulation from seeding to mid-tillering stages, but it presented the lowest values from heading to maturity stages and the lowest grain yield for no sufficient N supply at the reproductive stage. The grain yield of D50 was the highest, however, no significant difference was observed in grain yield, N agronomic efficiency or N recovery efficiency between D70 and D50, or between D70 and B50, while D70 was more labor saving than D50 for only one topdressing was applied in D70 compared with twice in other treatments. The above results indicated that 70% of fertilizer-N deeply placed as a basal fertilizer and 30% of fertilizer-N topdressed as a panicle fertilizer constituted an ideal approach for direct-seeded rice. This recommendation was further verified through on-farm demonstration experiments in 2015, in which D70 produced in similar grain yields as B50 did.