基于深度学习芒果图像在线识别与计数方法研究

为实现果树产量的智能评估,本研究对自然环境下的圣心芒果树图像进行果实识别研究,提出基于深度学习算法的芒果图像在线识别计数方法。首先,采用Faster R-CNN深度学习模型构建芒果图像识别算法;接着基于微信小程序与网页平台开发芒果图像上传模块,实现随时随地上传芒果图像至服务器;然后采用基于TCP协议的服务器客户端通信模式,并结合基于MATLAB平台的Faster R-CNN程序集,构建上传图像的在线分析模块,实现线上芒果图像的实时识别与计数;最终,芒果图像的识别与计数结果通过微信小程序和Web页面程序反馈给用户,内容包括单张图片和1个果园区域内所有图片的识别与计数结果,并实现青色芒果和红色芒果的分类统计。应用本研究构建的在线深度学习识别计数方法,在自然环境下采集125幅芒果图像进行测试试验。结果表明：芒果图像识别算法的计数识别准确率达到82.3%,其中漏检率与误检率分别为11.7%和8.6%,平均计数误差与计数误差率分别为4.2和7.9%;芒果图像在线识别计数方法能有效实现果树图像的采集、上传、识别与计数、分类统计和结果反馈,对整个果园区域内结果数量进行统计与分析,为果园的智慧管理提供科学决策依据。     

结合改进胶囊网络与知识蒸馏的茶青分类方法研究

不同等级茶青的准确分类,对名优茶产业发展至关重要,采用传统感官审评方法进行分选会使结果存在一定的主观性。采集茶青图像建立数据集,结合幽灵注意力瓶颈层与胶囊网络提出一种新型网络模型：GA-CapsNet。通过基于线性衰减比例系数的成长知识蒸馏方法对该模型进行训练,在迁移教师模型参数矩阵的同时,使学生模型随着迭代自适应降低依赖程度。试验结果表明,对比其他同类算法,所提出的方法在小规模数据集上分类性能优异,精确率、召回率及F1-score分别为94.97%、95.51%、95.24%。本研究基于机器视觉与深度学习技术构建了一种GA-CapsNet模型,为解决茶青分类问题提供了一种新思路。     

基于学习矢量量化神经网络的水稻白穗和正常穗的高光谱识别

水稻病虫害的发生会导致大量白穗的出现,对白穗和正常穗的区分是采取植保措施和灾害评估的基础。通过研究获取了由水稻二化螟和穗瘟造成的白穗和正常穗的室内光谱,选取红边斜率、红边面积、绿峰幅值和绿峰面积等4个高光谱变量作为输入向量,利用学习矢量量化（LVQ）神经网络对水稻白穗和正常穗进行分类。利用测试样本对网络进行测试,结果显示对白穗和正常稻穗的分类精度高达100%。研究表明,基于LVQ神经网络对水稻白穗和正常穗进行辨别的方法是切实可行的,可以补充和替代肉眼观测。     

基于多结构神经网络的大米外观品质评判方法

 应用多结构神经网络建立了大米外观品质评判模型,可实现5类大米的识别。模型采用5个并行工作的多层前向神经网络。每个多层前向神经网络包含两个隐含层,以大米图像的形状特征和颜色特征作为网络输入。网络训练和仿真结果显示模型识别的平均准确率为92.66%,比相同网络复杂度下的多层前向神经网络模型提高5.04个百分点,并且网络学习速率快。     

毯苗机插稻大面积应用侧深施肥的试验示范

通过毯苗栽插侧深施肥一体机进行大面积水稻侧深施肥示范,共设置8个处理,旨在探明水稻侧深施肥方式的增产效果和存在问题,提出今后机插稻合理的施肥方法。结果表明,采用毯苗机插侧深施肥平均可减少施用总氮量、总磷量分别约21.3%、26.4%,减少施肥次数1~2次;侧深施肥时水稻分蘖发生起步较早,具有明显的早发优势,较对照早发棵3~5 d;侧深施肥处理能增加水稻有效穗数、穗粒数、结实率,而千粒质量表现不一。     

Water supply from pearl millet by hydraulic lift can mitigate drought stress and improve productivity of rice by the close mixed planting

The authors have proposed the close mixed planting technique using mixed seedlings of two different crop species that results in close tangling of their root systems. Especially, the combination of drought-adaptive upland crops (e.g. pearl millet or sorghum) and flood-adaptive lowland crop of rice would be beneficial to overcome the drought and flood conditions and to reduce the risks of crop failure. In our previous studies, we found that upland crop yield losses by flood stress was mitigated by mix-cropped rice, owing to the oxygen gas released from the rice roots into the aqueous rhizosphere. In the present study, we conducted two experiments to assess whether mixed cropping a drought-resistant cereal, pearl millet, would improve the performance of co-growing drought-susceptible crop, rice under drought conditions. In the field experiment, some grains were obtained from the rice plants mix-cropped with pearl millet under drought condition. However, no rice matured in the single cropping system. In the model experiment using deuterium analysis, it was confirmed that water absorbed by pearl millet roots from deep soil layer was utilized by rice, suggesting that mix-cropped rice could withstand drought stress and complete grain filling using water released into the upper soil layer by hydraulic lift.     

基于图像处理技术和神经网络实现机采茶分级

为解决名优绿茶采摘环节的瓶颈问题,提出对机采大宗绿茶进行分级的思路。现有绿茶机采设备采摘的鲜叶一般只能制作普通的大宗绿茶,鲜叶存在混杂、破碎率高和老梗叶等问题,本文基于Labview vision、图像处理技术和神经网络算法分析机采绿茶成品的凸包面积、凸包周长、长轴长度、短轴长度等形态特征并对样本进行分类,实现从机采大宗绿茶中分选出名优绿茶。其中样本的形态特征采用工业CCD摄像头获取;用户界面用Labview自定义开发设计,数据交互方便,开发周期短。茶叶样本试验结果表明:该方案机采绿茶成品的分级正确率可以稳定在90%以上。本研究为进一步研究机采茶分级设备提供了良好的理论基础。     

Quality assessment of durum wheat storage centres in Sicily: Evaluation of vitreous, starchy and shrunken kernels using an image analysis system

The proportion of vitreous kernels in a sample is an internationally recognized specification for determining the value of durum wheat (Triticum durum Desf.). Vitreous kernels are mostly related to quality, which affects the pasta performance during cooking. Vitreousness and the amount of shrunken kernels are visually assessed during the grading process. This assessment is subjective and tedious.A machine vision system was developed to determine the percentage of vitreous, starchy, piebald and shrunken kernels in approximately 100 grain samples, using a trans-illuminated image of one layer of non-singulated kernels (in bulk) acquired by a digital camera. Classification models were developed with stepwise Linear Discriminant Analysis, as well as an on-line Bayesian classifier integrated with an image analysis system. The overall correct classification in Starchy classifier was high 98.58% in the Training set, made up of 6679 grains, following the Linear Discriminant Analysis classification, of 30 Italian cultivars harvested in 2005 in three localities. An independent Test set was constituted by samples collected in 30 Sicilian Storage Centres in the 2007 harvest season. The overall classification was 96.03%. For the Shrunken classifier 95.27% of the Training set and 99.58% of the Test set were correctly classified. The image analysis system was more reliable than the human inspectors who validated the system, both for the same samples measured many times and at different times.     

基于霍特林变换的稻米彩色图像粒形检测算法

利用稻米分割后轮廓灰度图与背景灰度图的灰度均值之差和灰度方差之差进行米粒图像分割效果定量评价,对7个彩色通道的稻米图像进行分割评判,选取I1(红色、绿色、蓝色通道的平均值)通道进行稻米图像分割。提取分割后标注的单粒米粒边界的二维坐标向量,对坐标向量进行霍特林变换,通过计算变换后米粒最小外接矩阵来表征稻米粒形,简化了现有的稻米粒形检测算法。检测稻米粒型时,算法在MATLAB7.5.0环境下运行。该算法所得米粒长宽比与人工检测结果的平均相对误差为1.65%,每幅图像平均耗时0.323s;而最小外接矩形算法的平均相对误差为2.24%,每幅图像平均耗时2·837s。     

印度稻米生产与标准研究

印度稻米等级标准源于1937年颁布的农产品分级和商标法。印度的大米标准分成常规稻米标准和巴斯马蒂米标准两大类。常规稻米分等标准相对简单,主要以含碎米率为分级指标,适用于国内贸易。巴斯马蒂米等级标准比较完善,以粒长、长宽比、不完善粒以及膨胀率等指标分等,专门用于出口贸易。总体来说,印度稻米标准起点不高,对碾磨程度、缺陷粒的要求较松,国际贸易中一般采用国际食品法典Codex的农残标准。     

印度稻米生产与标准研究

印度稻米等级标准源于1937年颁布的农产品分级和商标法。印度的大米标准分成常规稻米标准和巴斯马蒂米标准两大类。常规稻米分等标准相对简单,主要以含碎米率为分级指标,适用于国内贸易。巴斯马蒂米等级标准比较完善,以粒长、长宽比、不完善粒以及膨胀率等指标分等,专门用于出口贸易。总体来说,印度稻米标准起点不高,对碾磨程度、缺陷粒的要求较松,国际贸易中一般采用国际食品法典Codex的农残标准。     

Extractability and chromatographic separation of rice endosperm proteins

The molecular weight (MW) distribution of proteins extracted with different solvents from defatted rice endosperm was examined by size exclusion-high performance liquid chromatography (SE-HPLC) with 2.0% sodium dodecyl sulfate (SDS) (w/v) as mobile phase. The resulting protein peaks were further characterized by SDS-PAGE. Under the experimental conditions, 2.0% SDS extracted 64% of the proteins. Adding 6.0 M urea resulted in a 15% increase in extractability (up to 79%). With using 20–100 mM NaOH, 70–81% of the proteins were extractable. Maximum extractability was reached with 2.0% SDS, 6.0 M urea and 0.5–1.5% dithiothreitol (DTT). Apparent MW profiles of rice endosperm proteins allowed classification into six fractions of decreasing apparent MW. Fraction VI contained the low MW albumin, globulin, and prolamin protein material. Fractions IV and V originated from α and β glutelin subunits, respectively. The polypeptides of fraction III consisted of an α and a β subunit linked by an intermolecular disulfide bond. The polypeptides of fractions I and II were dimers, trimers or more highly polymerized forms of the (α–β) glutelin subunit dimer in fraction III. While the work confirmed that rice glutelin is composed of polymers of α and β subunits, remarkably, higher MW glutelin aggregates (fractions I–III) only partly dissociated on reduction. Low MW protein material (fraction VI) was entrapped in the aggregated protein network and was released on reduction. The rapid and reproducible SE-HPLC method developed for rice protein separation allows a more quantitative approach than SDS-PAGE.     

Type correlation analysis between six morphological trait index and eight isozyme loci

Indica-japonica differentiation is the majority for differentiation of Asian cultivated rice(O. Sativa L.). Sun et al proposed to distinguish indica and japonica by using the parameter Pj value which was quantified from six isozyme loci associated with indica and japonica differentiation, and the classification was consistent with the method of Cheng's "six combined morphological trait index" (CMT index). In this study, we analyzed the correlation between the six morphological traits and eight isozyme markers for indica-japonica classification by using 100 rice lines.     

粳稻品种的粒厚特征及其对食味品质的影响

 以2006年收获的71份北方粳稻主栽品种为试材,研究了糙米粒厚的分布及粒厚与食味品质的相关性。结果表明,品种间糙米粒厚差异较大,变异系数为6.57%,粒厚在1.81~2.10 mm的品种占参试品种总数的78.88％。糙米粒厚与米饭食味评分呈极显著正相关。用粒厚分级机对其中5个粳稻品种进行粒厚分级,测定同一品种不同粒厚样品的米饭食味评分值、蛋白质含量、直链淀粉含量和RVA谱特征值。结果表明,粒厚对稻米蛋白质含量、RVA特征值、米饭食味评分等指标有显著影响,同一品种随着粒厚的增加,食味品质得到改善。讨论了粒厚作为品种选育指标的必要性。     

黑米花色苷调节脂质代谢作用及其分子机制研究进展

近年来大量体内外实验研究证实花色苷具有调节脂代谢作用。黑米种皮富含花色苷,是天然花色苷重要的来源之一。本文综述了黑米花色苷对脂质代谢的调节作用,并分别从调节脂肪酸和胆固醇代谢相关酶及其基因的表达的角度总结了其作用机制,分析了研究中存在的主要问题,提出了下一步深入研究的重要方向,旨在为黑米精深加工和功能食品开发提供理论依据。     

Mapping rice-planted areas using time-series synthetic aperture radar data for the Asia-RiCE activity

In Asia, rice is a staple cereal crop and the continent accounts for about 90 % of the global rice production and consumption. Statistics on the areas planted with rice or production of paddy rice are fundamental to agriculture-related decisions or policy-making. Asia-Rice Crop Estimation & Monitoring (Asia-RiCE) aims to develop rice-related information, such as paddy field maps, rice growing conditions, yield, and production, using remote sensing tools and disseminate the same at the local and global scales. In this paper, we propose a methodology for the identification of rice-planted areas by using multi-temporal SAR images; a software named INternational Asian Harvest mOnitoring system for Rice (INAHOR) was developed to manipulate the proposed algorithm. The INAHOR uses the imagery observed both at the time of planting of rice and grown-up stages. In this study, two thresholds needed for the INAHOR were optimized based on the detailed land cover data collected through a field survey. Rice-planted areas across the study area in Japan were identified by the INAHOR using the RADARSAT-2 Wide Fine beam mode data. The classification results of RADARSAT-2 VV and VH polarizations were compared. The data with VH polarization showed a higher total accuracy of 83 % with ?20.5 dB and 3.0 dB for the minimum and range thresholds, respectively. The INAHOR is currently being used with the RADARSAT-2, ALOS, and ALOS-2 SAR data in the Southeast Asian countries to assess the robustness of the thresholds and classification accuracies under the framework of Asia-RiCE.     

基于ResNet卷积神经网络的绿茶种类识别模型构建

绿茶是我国种类最多、产量最大的茶类,外形是其分类的重要依据。图像分类是计算机视觉的核心技术之一,但其在茶叶领域的应用较少,茶类识别仍依赖感官审评方法。采集8种常见绿茶（丽水香茶、信阳毛尖、六安瓜片、太平猴魁、安吉白茶、碧螺春、竹叶青和龙井）共1 713张图片,基于ResNet卷积神经网络,从识别模型的预测能力、收敛速度、模型大小和识别均衡性等角度探索了不同网络深度和不同优化算法的建模效果,最终选择ResNet-18结构、SGD优化算法,建立了区分8种绿茶的深度学习模型,其对复杂背景茶叶图片的识别准确率达到了90.99%,单张图片识别时间仅为0.098 s,模型大小为43.7 MB。本研究为构建茶叶视觉识别模型并应用于移动端提供了基础,为茶叶种类识别提供了一种快捷而高效的新方法。     

Assessment of irrigation practices at the tertiary canal level in an improved system—a case study of Wasat area, the Nile Delta

Egypt faces great challenges due to its limited water resources by enforcement policies to improve the performance of the existing delivery system and its development. The improvement of irrigation systems in the Nile Delta is one of the most important attempts in Egypt to implement more effective irrigation technologies. This study was carried out to evaluate improved tertiary canal level and farmers’ practices by comparing with other unimproved systems to understand the farmers’ practices in their farms after modifying the existing irrigation system. This study area applied to the Wasat command area’s most commonly used to the cultivation of a paddy field in Egypt, which contributes 40 % of production. The overall results indicate that the water-use application at the improved system level improved. This was due to the role of water user association in the successful management and operation of the water-supply system on the private level of water distribution network. So, water users’ association has the positive effect on managing of the improved tertiary canal. Although, there are main problems of water delivery in the irrigation networks that was a water shortage in the main canal owing to its location at the tail of the feeder canal system in the Nile Delta, and other reasons include the absence of crop production planning by farmers, especially rice farmers in summer, and the greater demand of some fields than supply.     

水稻低丰度表达基因OsAMT1;3实时荧光定量PCR方法的建立及其应用

采用实时荧光定量PCR技术,通过使用特异引物,对水稻中的低丰度表达基因OsAMT1;3进行了转录水平上的定量分析,成功建立了可检测低丰度表达基因的SYBR Green实时荧光定量PCR技术平台。该方法具有很好的准确性和实用性。获得的荧光定量PCR扩增曲线,基线平整,指数区明显,斜率大且固定;线性范围广,17～36个循环都能测出;稳定性、重复性好,变异系数仅为0.47%;标准曲线表明,循环阈值与PCR体系中起始模板量的对数值之间有着良好的线性关系,可对基因表达进行相对定量;缺氮条件下OsAMT1;3 与纯NH4+处理相比表达量增加4倍以上。     

基于GF-1影像面向对象分类方法的水稻种植信息提取研究

应用遥感技术提取水稻种植信息是农业遥感的重要内容。GF-1卫星WFV数据为农业信息提取提供了新的途径,面向对象的分类方法是遥感解译的重要方法。本研究以扬州市为研究区域,基于GF-1影像WFV数据,采用面向对象的分类方法,提取水稻种植信息,并实地调查验证试验结果,试图探讨GF-1数据面向对象分类方法在水稻种植信息提取中的可行性与影响提取精度的因素。结果表明,应用GF-1数据,采用面向对象的分类方法能够很好地完成扬州市水稻种植信息的提取,2016年扬州市有水稻种植面积214 524 hm~2,总体精度达到98.5%,Kappa系数0.95,面积精度达97.5%;实地考察能够提高提取精度,地形破碎程度越低,提取精度越高。