基于原型网络的小样本禽蛋图像特征检测方法

机器视觉因具有检测速度快、稳定性高及成本低等优点，已发展成为禽蛋无损检测领域主流检测手段。使用该技术对禽蛋进行无损检测时，需要依赖大量禽蛋图像作为数据支撑才能取得较好的检测效果。由于养殖安全等限制，禽蛋图像数据的采集成本较高，针对该问题，提出了一种适应于小样本禽蛋图像检测的原型网络（Prototypical network）。该网络利用引入注意力机制的逆残差结构搭建的卷积神经网络将不同类别的禽蛋图像映射至嵌入空间，并利用欧氏距离度量测试禽蛋图像在嵌入空间的类别，从而完成禽蛋图像的分类。本文利用该网络分别验证了小样本条件下受精蛋与无精蛋、双黄蛋与单黄蛋及裂纹蛋与正常蛋的分类检测效果，其检测精度分别为95%、98%、88%。试验结果表明本文方法能够有效地解决禽蛋图像检测中样本不足的问题，为禽蛋图像无损检测研究提供了新的思路。     

基于距离相关系数和KNN回归模型的森林蓄积量估测研究

【目的】探究Landsat8 OLI数据和KNN算法在森林蓄积量估测中的潜力。【方法】以湖南省湘潭县为研究区,采用Landsat8 OLI数据和同时期的二类调查数据,通过距离相关系数筛选特征,分别采用线性回归模型(MLR)、K-近邻模型(KNN)、距离加权KNN模型(DW-KNN)和优化欧式KNN模型(FW-KNN)对森林蓄积量进行估测。使用十折交叉方法进行精度检验,对检验结果进行对比分析。【结果】3种KNN模型的估测结果均高于传统的线性模型,并且在3种KNN模型中,FW-KNN算法效果最好,决定系数达到0.69,为3种模型中最高;3种KNN模型中,本研究优化欧氏距离KNN模型的估测精度最高,其均方根误差为30.3%,相比于传统KNN模型的均方根误差降低了5.1%,相比于DW-KNN模型降低了3.3%。【结论】采用DW-KNN蓄积量估测结果明显优于其他两种模型,说明通过特征与蓄积量的相关性优化样本间的距离是一种可行的KNN优化方法。     

基于Web电子商务平台的食用菌产品共词聚类算法

针对食用菌产品电子商务平台的数据挖掘问题,采用了共词聚类算法进行主题聚类,在密度聚类算法的基础上,使用了移动非密集单元的处理方法实现了共词聚类。该算法克服了传统聚类算法裁剪造成的精度误差,提高了聚类结果的精确度,应用在电子商务平台的效果良好。     

基于空间分析的黑龙江稻瘟病重点发生区域识别

黑龙江是北方重要稻作区,稻瘟病是影响水稻产量的主要病害之一。依据稻瘟病测报调查规范,基于近5年黑龙江稻瘟病乡村监测点调查数据,利用空间分析技术的方法,分析黑龙江稻瘟病重点发生区县和核心发生区域,提取稻瘟病重点发生乡村监测点。结果表明,近5年黑龙江稻瘟病发病面积占调查面积的7.3%,发病等级以1级为主,整体发生较轻,稻瘟病重点发生区域主要分布在松花江上游地区哈尔滨市的阿城、尚志、方正、延寿、通河、依兰沿线和松花江下游三江平原地区佳木斯市的富锦、同江一带,以及嫩江流域大庆市的肇源县和齐齐哈尔市的讷河市以及呼兰河流域绥化市北林区,并提取了98个叶瘟和47个穗颈瘟重点监测点。上述结果为黑龙江稻瘟病调查监测区域和监测点的布局优化提供了重要依据,为稻瘟病测报防控工作提供重要的支撑。     

Regional and temporal differentiation of virulence phenotypes of Puccinia triticina from common wheat in Russia during the period 2001–2018

Leaf rust, caused by the fungus Puccinia triticina, is one of the most damaging rust diseases of wheat in Russia. Populations of P. triticina were monitored in seven regions of Russia from 2001 to 2018, with a total of 5,191 single urediniospore isolates from bread wheat (Triticum aestivum) being analysed. Populations have changed significantly in all regions since 2012, after 2 years of drought (2010–2011). Regional collections of P. triticina were also significantly different between the two periods 2001–2009 and 2012–2018, with changes along two geographic gradients from West Siberia to the north-west and south-west (North Caucasia) of the European part of Russia. All tested isolates were avirulent to resistance gene Lr9 in 2001–2009 but, since 2010, virulence to Lr9 has occurred and annually increased in the Asian part of Russia (Ural and West Siberia) due to deployment of cultivars with the Lr9 gene. Virulence to Lr2a and Lr15 was considerably lower in Dagestan (6%–33%) and all European regions (35%–67%) than in Asian regions (84%–96%). During 2001–2009, virulence on Lr1 was also lower in Dagestan (33%) and the European regions (50%–77%) than in Asia (91%–96%); however, by 2012–2018, nearly all isolates were virulent on Lr1. Remarkable changes were observed in frequencies of P. triticina races defined by their virulence/avirulence to Lr1 and Lr2a genes. We postulate the P. triticina population in Dagestan is specific to that area and pathogen populations in European and Asian parts of Russia are distinct.     

Ecological and landscape effects on genetic distance in an assemblage of headwater fishes

Environmental divergence along hierarchically structured longitudinal gradients may constitute barriers to gene flow in river networks for headwater specialised species. While known, this phenomenon has not been well studied, especially with regard to degree of headwater specialisation. We examined six headwater species that differ in habitat specialisation to assess whether patterns of differentiation vary according to geographic or environmental distance. We also identified regional environmental or anthropogenically induced fragmentation effects by comparing within‐drainage patterns of genetic distance across replicate watersheds. We used a comparative modelling framework to determine whether isolation by distance or isolation by resistance of large river habitats was a better predictor of genetic distance across species. The influence of reservoir presence and regional network characteristics that may influence the hydrology and size of large river habitats were also assessed. Resistance effects from large rivers were closely related to headwater specialisation, with increased specialisation leading to increased resistance and loss of drainagewide population connectivity. These results affirm that dendritic networks naturally fragment headwater specialised species. Further isolation from anthropogenic fragmentation was detected in two of the six drainages, indicating interactions with system‐specific conditions. Landscape variables related to the hydrology of large rivers also affected genetic distance in predicted ways, supporting the importance of large rivers in genetically structuring headwater species in drainage networks.     

空气罐与超压泄压阀联合水锤防护特性

提出了一种空气罐与超压泄压阀联合设置的停泵水锤防护方案,并基于瞬变流计算的特征线法建立了空气罐、液控蝶阀、超压泄压阀等边界条件的数学模型,模拟了停泵工况下系统的压力变化过程.结合工程实例,对比分析了空气罐单独防护、空气罐与两阶段关闭泵后液控蝶阀联合防护以及空气罐与超压泄压阀联合防护对停泵水锤的影响.在空气罐体型一定时,对超压泄压阀的启闭规律进行了敏感性分析.计算结果表明,空气罐与两阶段关闭泵后液控蝶阀联合防护方案对输水系统正负水锤防护均不利;而空气罐与超压泄压阀联合防护方案对输水系统正负水锤均有较好的防护效果,与空气罐单独防护方案相比,泵后高压管段最高压应力由1.343 MPa降至1.087 MPa,在满足承压标准1.2 MPa的基础上安全裕度提高了9.4%,空气罐体型也由200 m³缩减至160 m³.超压泄压阀应在5 s内开启至全开度,且开启后持续时间应接近1个相长.     

基于配合力和遗传距离的甜高粱杂种优势预测

【目的】对甜高粱主要农艺性状进行杂种优势、一般配合力及特殊配合力分析,同时,分析配合力、表型遗传距离以及分子遗传距离用于杂种优势预测的可行性,为甜高粱的种质创新和杂交种选育提供理论参考。【方法】采用不完全双列杂交设计,以8个甜高粱不育系为母本及8个甜高粱恢复系为父本配制64个杂交组合。对亲本及杂交后代进行2年的性状调查,包括：出苗至开花日数、生育期、株高、穗长、茎粗、分蘖、单穗粒重、千粒重、籽粒产量、单株重、生物产量和含糖量。分析不同性状的杂种优势、一般配合力、特殊配合力、表型遗传距离、分子遗传距离以及配合力、遗传距离与杂种优势的相关性。【结果】各性状的中亲优势由强到弱分别为单株重、籽粒产量、单穗粒重、生物产量、株高、穗长、千粒重、茎粗、生育期、至开花日数、分蘖和含糖量,其中,生育期、至开花日数、分蘖和含糖量等性状为负优势。不同性状的中亲优势和超亲优势由强到弱的顺序基本相同。配合力分析表明,每个性状中,不同亲本的一般配合力相差较大,且不同组合的特殊配合力也有很大差异。大多数特殊配合力高的组合,其亲本的一般配合力也较高。杂种优势与配合力和遗传距离的相关性为单株重、籽粒产量、单穗粒重、生物产量、穗长、千粒重、分蘖以及含糖量等性状的杂种优势与其亲本的一般配合力和特殊配合力均为极显著正相关。生育期的杂种优势与特殊配合力为极显著正相关,至开花日数与特殊配合力为显著正相关。亲本间的表型遗传距离为2.86—6.82,分子遗传距离为0.50—0.96。单株重、籽粒产量、单穗粒重、生物产量、株高、穗长、茎粗及含糖量等性状的杂种优势与分子遗传距离的相关性大于表型遗传距离,其中,生物产量、单株重、穗长和茎粗的杂种优势与分子遗传距离为极显著正相关。【结论】所有性状中,与产量相关性状的杂种优势较高,而含糖量和分蘖的杂种优势较低。在杂种优势预测上,利用亲本的配合力可有效预测杂种优势,预测效果优于遗传距离。与表型遗传距离相比,分子遗传距离对杂种优势的预测更有效。     

基于逆运动学降维求解与YOLO v4的果实采摘系统研究

为提高采摘设备的执行效率，采用六自由度机械臂、树莓派、Android手机端和服务器设计了一种智能果实采摘系统，该系统可自动识别不同种类的水果，并实现自动采摘，可通过手机端远程控制采摘设备的起始和停止，并远程查看实时采摘视频。提出通过降低自由度和使用二维坐标系来实现三维坐标系中机械臂逆运动学的求解过程，从而避免了大量的矩阵运算，使机械臂逆运动学求解过程更加简捷。利用Matlab中的Robotic Toolbox进行机械臂三维建模仿真，验证了降维求解的可行性。在果实采摘流程中，为了使机械臂运动轨迹更加稳定与协调，采用五项式插值法对机械臂进行运动轨迹规划控制。基于Darknet深度学习框架的YOLO v4目标检测识别算法进行果实目标检测和像素定位，在Ubuntu 19.10操作系统中使用2000幅图像作为训练集，分别对不同种类的果实进行识别模型训练，在GPU环境下进行测试，结果表明，每种果实识别的准确率均在94%以上，单次果实采摘的时间约为17s。经过实际测试，该系统具有良好的稳定性、实时性以及对果实采摘的准确性。