采用双模态联合表征学习方法识别作物病害

基于深度卷积神经网络的视觉识别方法在病害诊断中表现出色,逐渐成为了研究热点。但是,基于深度卷积神经网络建立的视觉识别模型通常只利用了图像模态的数据,导致模型的识别准确率和鲁棒性,都依赖训练数据集的规模和标注的质量。构建开放环境下大规模的病害数据集并完成高质量的标注,通常需要付出巨大的经济和技术代价,限制了基于深度卷积神经网络的视觉识别方法在实际应用中的推广。该研究提出了一种基于图像与文本双模态联合表征学习的开放环境下作物病害识别模型(bimodalNet)。该模型在病害图像模态的基础上,进行了病害文本模态信息的嵌入,利用两种模态病害信息间的相关性和互补性,实现了病害特征的联合表征学习。最终bimodalNet在较小的数据集上取得了优于单纯的图像模态模型和文本模态模型的效果,最优模型组合在测试集的准确率、精确率、灵敏度、特异性和F1值分别为99.47%、98.51%、98.61%、99.68%和98.51%。该研究证明了利用病害图像和病害文本的双模态表征学习是解决开放环境下作物病害识别的有效方法。     

基于迁移学习的GoogLeNet果园病虫害识别与分级

为提高果树病虫害危害程度分级精度进而更好地指导果园病虫害防治，采用迁移学习技术与GoogLeNet模型相结合的方法，对6种果园作物的25类病虫害样本进行识别与危害程度分级研究；同时，探究不同数据集大小以及不同优化算法对模型性能的影响；基于MATLAB平台设计了一款可视化的病虫害识别与分级系统。结果表明:1)基于迁移学习的GoogLeNet模型，对病虫害识别精度可达99.35%，危害程度分级精度可达92.78%；2)在相同训练参数下，本研究模型比AlexNet、VGG-16、ResNet-18、SqueezeNet、原GoogLeNet及MobileNet-v2模型验证精度提高了2.38%~11.44%，并且收敛速度最快；3)本研究模型识别精度随着数据集的增大而提高；在3种优化算法中SGDM算法耗时最短且精度最高，更适合本研究模型。通过拍摄果树叶片病害区域图像，本研究设计的系统能够在0.43 s左右准确识别出果树种类、病害类型以及危害等级等信息。     

茶树植保机械及减量施药技术研究进展

农药防治是茶树病虫害综合防治的重要组成部分,其在病虫害突发或爆发时具有快速高效的防治优势。茶树叶片表面具有亲水性,常量施药会造成茶叶农残超标、生态环境破坏等问题,实现茶树减量施药是减少茶叶农残的有效手段。系统综述了茶树生物特性、茶树病虫害预测诊断及防治方法、茶树植保机械及施药技术,强调提高茶树低容量喷雾的农药有效利用率是实现茶树减量施药的关键。针对目前茶园地面工况复杂及农药利用率低的问题,本文从低容量仿形喷雾机、茶树病虫害喷雾决策及智能终端等六个方面提出茶树病虫害施药技术及装备的研究建议,指出低量化、精准化及智能化是未来茶树植保喷雾机械及施药技术的发展方向。     

南充地区晚熟柑桔主要病虫害发生特点及综合防控措施

本文介绍了南充地区晚熟柑桔主要病虫害的种类及为害发生特点,并提出了相应的病虫害综合防控措施,保证晚熟柑桔的正常生长,以期为当地晚熟柑桔生产、种植户节本增收提供参考。     

Relationships Among the Feeding Behaviors of a Mirid Bug on Cotton Leaves of Different Ages and Plant Biochemical Substances

Apolygus lucorum (Meyer-Dür) (Hemiptera: Miridae) tends to feed on young plant tissues. To explore the relationship between stylet probing behaviors of adult A. lucorum and conditions of cotton leaves, we conducted an experiment using electropenetrography (EPG). Behaviors were recorded on four cotton varieties, in relation to thickness and biochemical traits of differently-aged leaves. Cotton leaf age had a significant effect on the probing behavior of A. lucorum but cotton variety did not. One-day-old leaves of A. lucorum received the highest mean number of stylet probes (penetrations) per insect, and longest mean durations per insect of combined stylet probing or its components, cell rupture and ingestion behaviors. All of the leaf traits (thickness and biochemical substances) were similar among these four cotton varieties. Leaf thickness had a significantly negative effect on the same four variables above. Gossypol and tannin also had a negative impact on combined probing duration. Redundancy analysis showed that the four EPG variables were closely related to nutrient substances (amino acids, sugar, and water) while they had the opposite relationship with plant defense substances (gossypol and tannin). On cotton in the seedling stages, A. lucorum fed more readily on the youngest, thinnest leaves in our no-choice EPG experiments. Nutrients and chemical resistance substances determined the probing duration of A. lucorum. Our findings can contribute to better understanding of patterns of feeding and host consumption by A. lucorum, ultimately improving cotton resistance to A. lucorum.     

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。     

马尾松叶表微生物多样性

为发掘对林木生长发育有利的优良微生物资源,并筛选适合叶表微生物的收集方法,以马尾松针叶为试验材料,分别用悬摇法和超声波法收集马尾松叶表微生物,用扩增子高通量测序技术、MUSCLE和Qiime软件研究马尾松叶表微生物的多样性。结果表明:扫描电镜观测结果显示,马尾松针叶表面定殖有大量微生物,包括真菌(菌丝及孢子)和细菌。扩增子高通量测序结果表明,马尾松叶表微生物物种丰富,包含细菌运算分类单位(OTUs)490个,真菌OTUs 1273个。马尾松叶表细菌以未分类的蓝细菌属(unidentified_Cyanobacteria)(36.53%)、未分类的拜叶林克氏菌属(unidentified_Beijerinckia)(28.60%)、鞘氨醇单胞菌属(Sphingomonas)(2.35%)为优势属;叶表真菌以枝孢属(Cladosporium)(2.45%)、拟盘多毛孢属(Pestalotiopsis)(0.92%)、无头孢菌属(Capnobotryella)(0.91%)为优势属。针对叶表细菌多样性的研究表明,悬摇法和超声波法均有较高的物种检出度;在叶表真菌多样性的研究中超声波法优于悬摇法,但超声波法样品间数据变异性较大,测定结果不稳定。     

Determination of critical nitrogen dilution curve based on leaf area index for winter wheat in the Guanzhong Plain,Northwest China

Excessive use of nitrogen(N) fertilizers in agricultural systems increases the cost of production and risk of environmental pollution. Therefore, determination of optimum N requirements for plant growth is necessary. Previous studies mostly established critical N dilution curves based on aboveground dry matter(DM) or leaf dry matter(LDM) and stem dry matter(SDM), to diagnose the N nutrition status of the whole plant. As these methods are time consuming, we investigated the more rapidly determined leaf area index(LAI) method to establish the critical nitrogen(N_c) dilution curve, and the curve was used to diagnose plant N status for winter wheat in Guanzhong Plain in Northwest China. Field experiments were conducted using four N fertilization levels(0, 105, 210 and 315 kg ha-1) applied to six wheat cultivars in the 2013–2014 and 2014–2015 growing seasons. LAI, DM, plant N concentration(PNC) and grain yield were determined. Data points from four cultivars were used for establishing the N_c curve and data points from the remaining two cultivars were used for validating the curve. The N_c dilution curve was validated for N-limiting and non-N-limiting growth conditions and there was good agreement between estimated and observed values. The N nutrition index(NNI) ranged from 0.41 to 1.25 and the accumulated plant N deficit(N_(and)) ranged from 60.38 to –17.92 kg ha~(-1) during the growing season. The relative grain yield was significantly affected by NNI and was adequately described with a parabolic function. The N_c curve based on LAI can be adopted as an alternative and more rapid approach to diagnose plant N status to support N fertilization decisions during the vegetative growth of winter wheat in Guanzhong Plain in Northwest China.     

北京地区南瓜病毒病病原种类鉴定

为明确北京地区南瓜病毒病种类及其主要侵染病原,2016~2017年在北京周边采集疑似感染病毒的南瓜病样84份,并根据南瓜上的6种病毒特异性引物对其进行反转录PCR(RT-PCR)检测。结果表明:共有79份南瓜病样检测显示阳性,其中黄瓜花叶病毒(Cucumber mosaic virus,CMV)的检出率最高,为52.38%;其次是小西葫芦黄花叶病毒(Zucchini yellow mosaic virus,ZYMV)和西瓜花叶病毒(Watermelon mosaic virus,WMV),检出率分别为44.01%、14.29%,其他病毒暂未检出。此外,16.67%的样品受2种病毒复合侵染,CMV和ZYMV复合侵染占7.14%,ZYMV和WMV复合侵染占9.52%。北京地区南瓜上优势病毒种类为CMV,且存在病毒复合侵染现象。