基于图像分块及重构的菠菜重叠叶片与杂草识别

针对重叠叶片在识别过程中存在识别率低、形状特征失效等问题,该研究提出一种基于图像分块及重构的方法,实现菠菜重叠叶片杂草识别。采用超绿模型将菠菜RGB图像进行灰度化得到绿色植被前景图像。针对重叠叶片形状特征失效问题,采用图像分块方式得到不同大小的图像块,并提取图像块中作物和杂草的颜色特征、局部二值模式(local binary pattern,LBP)纹理特征、分形盒维数共78维特征,构造支持向量机(support vector machine,SVM)分类器完成图像块的分类识别。该研究提出图像块边缘扩充和投票窗口机制得到重构图实现图像块分类结果可视化。试验结果表明,该研究提出的方法平均识别率达到83.78%,高于K最近邻法(K-Nearest neighbor,KNN)、决策树法等,可以实现重叠叶片的杂草识别,从而为智能除草机的研制提供理论依据。     

基于注意力机制及多尺度特征融合的番茄叶片缺素图像分类方法

针对番茄早期缺素性状不明显及各生长期特征差异较大所导致的特征区域尺寸不一致、难提取、难辩别等问题,提出了一种基于注意力机制及多尺度特征融合卷积神经网络的番茄叶片缺素图像分类方法（Multi-Scale Feature Fusion Convolutional Neural Networks Based On Atte ntion Mechanism,MSFF-AM-CNNs）。首先根据番茄叶片缺素特点提出了多尺度特征融合结构（Multi-Scale Feature Fusion Module,MSFF Module）;其次在DenseNet基础上,结合浅层网络主要提取纹理、细节特征,深层网络主要提取轮廓、形状特征的特点分别提出具有针对性的特征提取方法,通过不同形式引入注意力机制及多尺度特征融合结构,使全局多尺度信息融合多个特征通道、选择性地强调信息特征并达到对特征精准定位的功能;同时引入Focal Loss函数以减少易分类样本的权重。试验结果表明,MSFF-AM-CNNs的平均召回率、平均F1得分、平均准确率较原模型DenseNet-121均大幅提升,其中缺氮和缺钾叶片的准确率分别提高了8.06和6.14个百分点,召回率分别提高了6.31和5.00个百分点,F1得分分别提高了7.25和5.55个百分点,平均识别准确率可达95.92%,具有较高的识别准确率及广泛的适用性,能够满足番茄叶片缺素图像的高精度分类需求,可为植物叶片缺素识别提供参考。     

基于多光谱图像和数据挖掘的多特征杂草识别方法

为满足变量喷洒对杂草识别正确率的要求,提出一种基于多光谱图像和数据挖掘的杂草多特征识别方法。首先对多光谱成像仪获取的玉米与杂草图像从CIR转换到Lab颜色空间,用K-means聚类算法将图像分为土壤和绿色植物,随后用形态学处理提取出植物叶片图像,在此基础上提取叶片形状、纹理及分形维数3类特征,并基于C4.5算法对杂草分别进行单特征和多特征组合的分类识别。试验结果表明,多特征识别率比单特征识别率高,3类特征组合后的识别率最高达到96.3%。为验证该文提出方法的有效性,将C4.5算法与BP算法以及SVM算法进行比较,试验结果表明C4.5算法的平均识别率高于另2种算法,该文提出的田间杂草快速识别方法是有效可行的。该文为玉米苗期精确喷洒除草剂提供技术依据。     

基于水培生菜力学特征的成熟度分类方法

为了能准确识别生菜的成熟度,实现生菜适时采收,避免因采收期不当而造成品质下降等问题,该研究提出用穿刺试验力学特征表征水培生菜成熟度的方法,提取同一颗生菜不同叶片的不同部位时域和频域的力学特征,得到叶片力学特征与叶片成熟特性指标的相关性。为了对水培生菜未成熟株、成熟株、过成熟株进行准确分类,设计双阈值深度遍历算法,确定分类准确率最高的叶片类型和区域;采用6种机器学习算法以该区域所有的力学特征为输入,以成熟度3种分类为输出进行训练。试验结果表明,生菜叶片力学特征与成熟度特性指标紧密相关,生菜中叶叶茎区域分类准确率最高,可优化集成分类机器学习算法准确率最高为94.3%。研究结果提供了一种应用力学特征解决水培生菜成熟度检测与分类的新方法。     

利用融合高度与单目图像特征的支持向量机模型识别杂草

除草是保证农作物高产的必要工作。针对机械化除草和智能喷药中存在的杂草识别问题,以2~5叶苗期玉米及杂草为研究对象,进行了融合高度特征与单目图像特征的杂草识别方法研究。首先从单目图像中提取16个形态特征和2个纹理特征;然后基于双目图像,提出了针对植株的高度特征提取方法,所得高度特征与实际测量值间误差在±12 mm以内;利用max-min ant system算法对形态特征进行优化选择,将形态特征减少到6个,有效减少数据量62.5%,并与纹理和高度特征进行融合;将2~5叶玉米幼苗的可除草期划分为3个阶段,分别构建融合高度特征与单目图像特征的SVM识别模型,并与相应不含高度特征模型进行对比。经测试,3个阶段模型的识别准确率分别为96.67%,100%,98.33%;平均识别准确率达98.33%。不含高度特征模型的识别准确率分别为93.33%,91.67%,95%;平均识别准确率为93.33%。结果表明,融合高度特征与单目图像特征的SVM识别模型优于不含高度特征模型,平均识别准确率提高了5百分点。该方法实现了高准确率的杂草识别,研究结果为农业精确除草的发展提供参考。     

基于改进DenseNet的田间杂草识别

精确、快速地获取作物和杂草的类别信息是实现自动化除草作业的重要前提。为解决复杂环境下农作物田间杂草种类的高效准确识别问题,该研究提出一种基于改进DenseNet的杂草识别模型。首先,在DenseNet-121网络的基础上,通过在每个卷积层后引入ECA（Efficient Channel Attention）注意力机制,增加重要特征的权重,强化杂草特征并抑制背景特征;其次,通过DropBlock正则化随机隐藏杂草图像部分特征块,以提升模型的泛化能力,增强模型识别不同类型杂草的适应性;最后,以自然环境下玉米幼苗和6类伴生杂草作为样本,在相同试验条件下与VGG-16、ResNet-50和未改进的DenseNet-121模型进行对比试验。结果表明,改进的DenseNet模型性能最优,模型大小为26.55 MB,单张图像耗时0.23 s,平均识别准确率达到98.63%,较改进前模型的平均识别准确率提高了2.09%,且综合性能显著高于VGG-16、ResNet-50模型;同时,通过采用CAM（Class Activation Mapping）可视化热度图方法分析,得出改进前后模型的类别判断概率分别为0.68和0.98,本文模型明显高于未改进模型,进一步验证了改进模型的有效性。该模型能够很好地解决复杂环境下农作物和杂草的种类精准识别问题,为智能除草机器人开发奠定了坚实的技术基础。     

基于机器视觉的田间杂草识别技术研究进展

田间杂草识别技术是实现变量喷洒除草剂以保护环境的关键所在。针对国内外在精细农业的杂草识别领域，全面、系统地分析了基于机器视觉的田间杂草识别技术的研究进展与应用状况，以促进该项技术在中国的应用和发展。分别阐述了利用植物和背景形状特征、纹理特征、颜色特征和多光谱特征识别田间杂草技术的理论依据、特征参数、研究状况和问题所在，并指出了实现田间实时识别的难点。     

基于卷积神经网络提取多尺度分层特征识别玉米杂草

为提高作物与杂草识别的准确率、稳定性和实时性,该文以幼苗期玉米及杂草为研究对象,提出了基于卷积神经网络提取多尺度分层特征的玉米杂草识别方法。首先建立卷积神经网络模型,以从图像的高斯金字塔中提取多尺度分层特征作为识别依据,再与多层感知器相连接实现图像中各像素的识别;为了避免目标交叠所带来的问题,对图像进行超像素分割,通过计算每个超像素内部的平均像素类别分布确定该超像素块的类别,再将相同类别的相邻超像素合并,最终实现图像中各目标的识别。试验结果表明:该方法的平均目标识别准确率达98.92%,标准差为0.55%,识别单幅图像的平均耗时为1.68 s,采用GPU硬件加速后识别单幅图像的平均耗时缩短为0.72 s。该方法实现了精确、稳定和高效的玉米与杂草识别,研究可为精确除草的发展提供参考。     

番茄缺素叶片的图像特征提取和优化选择研究

在基于计算机视觉技术对无土栽培番茄营养元素缺乏智能识别研究中,对不易被肉眼判别的缺氮和缺钾初期叶片进行图像特征的提取和优化选择研究，以提高识别的准确率。提出了应用相对差值百分率直方图提取缺素叶片的颜色特征，应用差分算子提取纹理的时域特征、应用傅里叶变换提取纹理的频域特征、应用小波包提取纹理的时频特征等的新方法，并新提出从颜色和纹理时域、频域、时频域等多个角度集成提取缺素叶片图像的有效特征，利用遗传算法对提取的众多特征项进行优化选择，以使诊断识别用的信息分类能力最优。试验表明，该方法识别的准确率较高，达到95%～92.5%，而且可以比肉眼识别提前6～10 d。     

空洞卷积结合全局池化的卷积神经网络识别作物幼苗与杂草

针对传统Alex Net模型参数大、特征尺度单一的问题,该文提出一种空洞卷积与全局池化相结合的多尺度特征融合卷积神经网络识别模型。通过对初始卷积层的卷积核进行膨胀,以增大其感受野而不改变参数计算量,并采用全局池化代替传统的全连接层来减少模型的参数。通过设置不同膨胀系数的初始卷积层卷积核与全局池化层类型,以及设置不同Batch Size,得到8种改进模型,用于训练识别共12种农作物幼苗与杂草,并从建立的模型中选出最优模型。改进后的最优模型与传统Alex Net模型相比,仅经过4次训练迭代,就能达到90%以上的识别准确率,平均测试识别准确率达到98.80%,分类成功指数达到96.84%,模型内存需求减少为4.20 MB。实际田间预测野芥与雀麦幼苗的准确率都能达到75%左右,说明该文最优模型对正常情况下的幼苗识别性能较好,但对复杂黑暗背景下的甜菜幼苗准确率为60%,对恶劣背景下的识别性能还有待提升。由于模型使用了更宽的网络结构,增加了特征图的多尺度融合,保持对输入空间变换的不变性,故对正常情况下不同作物幼苗与杂草的识别能力较强。该文改进模型能达到较高的平均识别准确率及分类成功率,可为后续深入探索复杂田间背景下的杂草识别以及杂草与幼苗识别装置的研制打下基础。     

两种生态控草措施对丘陵茶园杂草群落及物种多样性的影响

实现丘陵茶园杂草生态控制, 可有效减少茶园土壤水分和养分的消耗, 减少或避免使用化学除草剂.本研究根据生态位先估和生态位竞争原理, 设计覆盖稻草和间种豆科绿肥白三叶草(Trifolium repens)两种杂草生态控制技术, 同时设置清耕对照(CK), 采用小样方多点取样连续两年调查茶园杂草种群数量、生物量、平均株高等指标, 发现茶园共有杂草16科31属31种; 3个处理春季(4月)发生的杂草特征表现为生长速度较慢、生物量小、株高在30 cm以下; 夏季(7月)和秋季(9月)对照的优势恶性杂草多度(Pi)高达0.788和0.759, 稻草覆盖处理分别为0.256和0.420, 间种白三叶草处理恶性杂草种类较少发生(Pi值均低于0.050); 稻草覆盖处理茶园春、夏和秋季杂草物种丰富度(S)、种群多样性(H)、均匀度(J)等指数均高于对照, 优势集中性指数(C)低于对照; 间种白三叶草处理春、夏和秋季杂草总数量和生物量均显著低于对照, 夏、秋季杂草平均株高显著低于对照.结果表明两种生态控草处理措施能改变茶园杂草群落结构, 有效控制杂草的生长.     

Effect of soil tillage and weed control methods on weed biomass and yield of lentil (Lens culinaris Medic.)

Abstract

The study was carried out in dryfarming areas in Ankara, Turkey, over 2 years (2001 – 2002 and 2002 – 2003). The objective was to determine different soil tillage and weed control methods on weed biomass and yield components, yield of lentil (Lens culinaris). This study compared the effects of two tillage systems (shallow minimum tillage and traditional tillage) and three weed control methods (weedy check, hand weeding and herbicide) on weed biomass, growth characteristics, seed yield and some yield components of lentil. Significant differences were found among weed control methods for weed biomass and yield parameters of lentil. Tillage systems had no significant effect on weed biomass or yield of lentil. The highest yield and lowest weed biomass was found in the hand-weeded treatment compared to the other weed control methods. Results of this research indicate that weeds are a main constraint for lentil growing under dryland conditions. Grain yield of lentil was reduced more than 60% due to uncontrolled weeds.     

杂草基因组学研究进展及策略

本文分析了杂草的特异性和基因组特点,列出杂草基因组学研究的候选模式杂草植物种.论述了分子作图、比较基因组学、RNAi、TILLING、T-DNA插入突变和基因芯片等技术在杂草基因组学研究中的应用现状,以期全面认识和解析控制杂草特异性状的遗传基础和调控机理.重点阐述了杂草基因组学研究进展及加快杂草基因组学研究的技术策略,指出开展杂草基因组学的研究有利于建立科学控制杂草技术体系,改良作物产量和品质,从而提高农业的可持续生产力.     

基于多特征的田间杂草识别方法

该文阐述了通过利用植物的多种特征实现田间杂草的精准自动识别的方法。该方法先利用颜色特征分割土壤背景,然后利用位置和纹理特征识别行间和行内杂草,最后利用形态特征后处理误识别的作物和杂草。在实验室内利用实地采集的3～5叶期、不同作物行数的麦田图像对该方法进行了测试。作物和杂草的正确识别率最低为89%,最高为98%;处理时间最低为157 ms,最高为252 ms。试验结果表明：基于多特征的田间杂草识别方法具有较高的识别率和较快的识别速度。     

Can conversion to organic farming restore the species composition of arable weed communities?

Benefits of organic cropping for biodiversity have usually been demonstrated in relation to modern conventional cropping. In this study, the ability of organic cropping to restore species composition of weed communities was explored by comparing weed communities of present day organic cropping with weed communities at the beginning of the application of modern cropping measures in the 1960s. The data of two weed surveys of spring cereals (conducted in 1961-1964 and 1997-1999) in Finland were utilized for a comparison. Frequency of occurrence and density (plants m⁻²) of 41 weed species were compared between decades. Partial canonical correspondence analysis (pCCA) was applied to explore the relationship between species composition, management and non-management variables. Eight species had lower and 30 species higher or similar frequency of occurrence in the 1990s’ organically cropped fields than in the fields of the 1960s. However, 18 species had lower and 20 species had higher or similar density in the organically cropped fields than in the fields of the 1960s. Three species were not detected at all in the organically cropped fields. Crop and under-sown grass explained more of the variation in species composition in the 1990s than in the 1960s. The role of drainage and pre-crop was more important in the 1960s than in the 1990s. The most immediate benefit was gained by nitrophilous species that had suffered from herbicide application. The recovery of perennials and non-nitrophilous species will take a longer time. The results suggest that despite some benefits for biodiversity, organic farming at early phase cannot recover weed populations to the same level as before application of intensive cropping measures.     

Precise tillage systems for enhanced non-chemical weed management

Soil and residue manipulation can assist weed management by killing weeds mechanically, interfering in weed lifecycles, facilitating operations and enhancing crop establishment and growth. Current tillage systems often compromise these functions, resulting in heavy reliance on herbicides, particularly in no-till systems. Herbicides are an exhaustible resource, so new approaches to merge soil conservation and non-chemical weed management are needed. This paper broadly reviews various preventive and curative non-chemical weed management tactics. It also demonstrates how innovations can be derived from functional requirements of weed management operations, and from biological processes and weaknesses in weed's lifecycles. Mechanical weeding and enhancement of weed seed mortality are highlighted as examples. Major limitations with mechanical weeding include limited weed control in crop rows at early vulnerable crop stages, weather-dependent effectiveness, and difficulties in handling crop residues. Precise steering and depth control, improved seedbed friability and lighter tractors or controlled traffic could bring considerable improvements. To expose weed seeds to predators, position them for fatal germination, viability loss or low emergence may require completely different soil displacement patterns than those of current implements and systems. Controlled traffic and precise strip tillage offer good opportunities for implementing these weed management strategies in minimum-tillage systems.     

Contribution of fallow weed incorporation to nitrogen supplying capacity of paddy soil under organic farming

ABSTRACT

We studied soil nitrogen (N) management in a farmer’s organic rice farming in Japan, where the farmer applied no external N but incorporated gramineous fallow weeds and rice residues as in situ N sources. We focused on the effect of fallow weed incorporation on N-supplying capacity of the paddy soil by tracking decomposition of ¹⁵N-labeled fallow weeds after incorporation. The result fits well to the first order kinetics with the decomposition rate of 34.3% a year. A model of soil N accumulation and mineralization based on the first order kinetics showed that soil organic N originated from the incorporated weed would become saturated at the level 1.92 folds the annual input of weed N after several consecutive years of the incorporation. Mineralizable soil N (Min-N) of the weed origin would also become saturated after several years accounting for 21.2% of the total Min-N which includes the indigenous soil N from plow layer. We suspended weed incorporation (SWI) in a sub-plot of the fields for two consecutive years to compare Min-N therein with that in another subplot in the same fields subjected to continued weed incorporation (CWI). After 2 years of the suspension treatment, Min-N in SWI decreased to a similar extent as estimated with the soil-N model based on the first-order kinetics, with which we estimated that 16.9% of annual N uptake by the rice plants originated from the weed including 5.9% from the weed incorporated in the same year and 11.0% from that in the past years. N inflow to soil organic N from the weed was very close to N outflow attaining the steady state. The rice yield could thus be sustained by maintaining the soil N-supplying capacity via the internal cycling of fallow weed N.     

Changes in the weed communities as affected by different primary soil tillage and deep loosening

Long-term soil cultivation at the same depth affects soil characteristics and crop productivity. The aim of the study was to investigate the impact of a long-term different intensity soil tillage methods and deep loosening on weed number, weed agrobiological group and soil seed bank changes in till Bathygleyic Dystric Glossic Retisol soil under the climatic conditions of the Western Lithuania (geographical coordinates 55°43′38″N, 21°27′43″E). The study included different soil tillage methods (conventional ploughing, shallow ploughing and shallow ploughless tillage) and deep loosening. During investigational years, the greatest weed number in crops and the greatest weed seed number in the seed bank were determined in the soil reduced tillage (shallow ploughing and shallow ploughless tillage). The weed number in crops of conventional ploughing soil was 35.8% lover compared to reduced tillage soil. The weed seed number in the seed bank of conventional ploughing was 49.6% lover compared to reduced tillage Decreasing soil tillage intensity resulted in weed seeds concentration in the upper topsoil. A one-time deep loosening had a significant effect during the crop rotation: the weed number in crops and weed seed number in the seed bank were determined to have increased by 26.6% and 51.6% in conventional ploughing soil and by 11.9% and 23.2% shallow ploughless soil respectively. However, after deep loosening, the number of Poa annua in crops decreased 2.9 times in plots of conventional ploughing and 1.7 times – in plots of shallow ploughing soil.     

基于处方图的室内变量喷药除草系统设计

为提高除草剂利用率,开发了一种基于处方图的变量对靶施药除草系统。该系统以S3C2410微处理器为核心,在linux操作系统环境下,能根据MATLAB软件对机器视觉系统采集的棉田杂草图像处理后生成的喷施除草剂处方图,控制不同喷头开关及实现4种喷药量。对棉花与杂草识别、杂草质心坐标及面积大小计算,网格划分、田间杂草定位以及喷头高度与间距关系等关键问题进行了研究。在室内试验台上对喷药系统进行了测试,当喷药系统压力为0.1 MPa,传送带速度为0.4 m/s时,50棵杂草对靶率超过90%,与传统喷药方式相比可节省除草剂用量60%以上。该喷药系统的设计,为实时变量对靶喷药除草机器人的研究提供了参考。     

规模化稻鸭生态种养对稻田杂草群落组成及物种多样性的影响

为探索规模稻鸭生态种养稻田杂草群落的变化动态,运用植物群落生态的方法研究了规模稻鸭生态种养、常规稻鸭生态种养和水稻单一种植稻田杂草的群落变化特点及控草效果。结果表明:规模稻鸭生态种养杂草密度显著降低,与水稻单一种植比较防效为94%;水稻分蘖到孕穗期稻田杂草主要由看麦娘(Alopecuruspratensis)、稗草(Echinochhloa crausgalli)、狗牙根(Cynodon dactylon)、水花生(Alternanthera philoxeroides)、鸭舌草(Mouochoria vaginalis)、水竹叶(Murdannia triquetra)和鳢肠(Eclipta prostrata)组成,抽穗期到成熟期主要由水花生、稗草、狗牙根和水竹叶组成。研究还表明,规模稻鸭生态种养水稻分蘖期物种丰富度、Simpon指数、Shannon-Wiener指数略高于水稻单一种植,Pielou指数低于水稻单一种植;孕穗期到成熟期物种丰富度、Simpon指数、Shannon-Wiener指数低于水稻单一种植,Pielou指数显著高于水稻单一种植。水稻全生育期规模稻鸭生态种养与常规稻鸭生态种养比较稻田杂草密度、杂草生物量和多样性指数差异不显著,但杂草生物多样性指数略有提高。说明规模稻鸭生态种养显著改变稻田杂草的群落结构和组成,可抑制杂草发生危害,达到有效控草目的。