基于机器视觉的核桃仁动态分级研究

【目的】基于机器视觉技术研究出一种适合新疆核桃仁动态分级处理的方法。【方法】利用实时采集且已经完成图像预处理的样品核桃图像得到核桃仁特征集合,运用mRMR特征选择算法筛选原始特征集并对特征的重要性进行排列,通过对支持向量机、决策树和朴素贝叶斯三种机器学习算法进行模型训练和测试,得出最佳分级方法,设计核桃仁自动追踪方法和动态分级流程,构建的核桃仁自动分级系统。【结果】在使用特征bin19、K1和bin15训练朴素贝叶斯分类器时,核桃仁的分级正确率达到最大为97.33%,在动态条件下运用构建的核桃仁自动分级系统对150个核桃仁进行分级测试,正确率为81.33%。【结论】基于机器视觉研究出的核桃仁特征提取与分级方法、核桃仁动态分级处理动作方法,可以有效完成对核桃颜色和完整度的分级。     

基于机器视觉技术的单粒葡萄质量与果径预测分级研究

【目的】采用机器视觉技术,针对新疆无核白和红提单粒葡萄的质量和果径大小进行预测和分级研究。【方法】在不同的颜色特征空间模型,预处理原始图像,采用最大类间方差法分割目标区域;采用数学形态学方法去除二值图像中部分果梗及噪声点,获得最佳二值图像;基于二值图像,分析获取单粒葡萄的几何特征;最后,分别采用一元线性回归法和偏最小二乘回归法预测单粒葡萄的质量和果径,采用二次判别分析法对单粒葡萄的质量和果径进行分级。【结果】利用短轴与果形指数特征相结合建立的偏最小二乘回归模型可有效预测单粒葡萄的质量和果径,预测决定系数达到0.98和0.945;基于该特征组合的二次判别分析法可用于单粒葡萄的质量和果径分级,准确率超过85%。【结论】机器视觉技术能够较准确预测单粒葡萄的质量和果径,并能对质量和果径进行分级。     

基于小波神经网络的柑橘pH机器视觉检测

 【目的】研究涟红温州蜜柑pH的机器视觉检测及影响检测精度的因素。【方法】对机器视觉系统采集的柑橘图像进行图像裁切、RGB空间至HSI空间的转换和差值法去图像背景，用色调H和饱和度S为输入，建立小波神经网络柑橘pH预测模型，无损检测柑橘pH。【结果】30个测试样本的检测结果表明，预测偏差最大值为9.95%、偏差最小值为-3.6%、平均偏差为0.8%、标准偏差为2.95%，pH±0.1精度内的正确识别率为80%，pH±0.2精度内的正确识别率为93.33%。【结论】涟红温州蜜柑pH与果皮色泽之间具有相关性，可用机器视觉检测其pH。但进一步提高预测精度，首先须在图像处理环节上去除各种虫斑与病斑的影响。     

基于近地光谱特征的玉米田间杂草识别研究

化学防治是我国农田杂草防治使用较广泛的方法之一,化学除草剂的过量喷洒以及粗略的施用方式已成为农药泛滥、质量安全问题的罪魁祸首。目前,精准施药技术成为杂草去除的重要手段,杂草识别又是精准施药的关键技术。利用ASD FieldSpec 4便携式地物光谱仪,采集玉米、马齿苋、野苋菜及香附植株冠层在350～2 500 nm波段内的光谱信息,经过数据预处理,运用逐步判别模型,筛选出了954、1 324、1 869、734 nm 4个特征波段。将特征波段带入贝叶斯判别函数模型,分别对玉米田间杂草进行预测。结果表明,贝叶斯判别函数模型正确识别率达85.8%;对玉米的识别精度达90.0%。特征波段选取中剔除了波长749 nm选入了734 nm波长变量,在"红边"680～780 nm区域的反射率对玉米田间杂草识别较为重要。试验结果进一步论证了基于贝叶斯判别模型方法的可靠性,且证明了高光谱在杂草的识别方向具有一定的应用价值,该研究结果为田间杂草识别及光谱传感器提供了参考。     

机器人摘棉花可以识别好与坏

近日,南京农业大学工学院副教授王玲团队成功破解了采摘机器人对于棉花品级视觉识别的关键技术一田间籽棉品级识别。一旦得到运用。采摘机器人便可根据我国籽棉品级文字标准，籽棉的大小、白度、黄度和杂质量等特性，迅速、准确地判断出籽棉的品级．解决我国棉花的采摘质量问题．亦将有利于符合国情的棉花定级仪器的早日问世．填补空白。     

基于机器视觉的核桃仁特征提取与分级方法研究

基于机器视觉技术研究出一种适合新疆核桃仁分级特征提取与分级的方法。该方法利用已经完成图像预处理的实时采集的样品核桃图像,运用OpenCV完成从RGB到HSV的颜色空间转换,提取核桃仁颜色和完整度特征,建立原始特征矩阵特征,利用mRMR特征选择算法筛选原始特征集并对特征的重要性进行排列,最后通过对支持向量机、决策树和朴素贝叶斯3种机器学习算法进行模型训练和测试,得出最佳分级方法。结果表明,在使用特征bin19、K_1和bin15训练朴素贝叶斯分类器时,核桃仁的分级正确率达到最大,为97.33%。故得出基于机器视觉研究出的核桃仁特征提取与分级方法可以完成对核桃颜色和完整度的分级任务的结论。     

基于行宽的玉米行间杂草识别算法

为精确识别和定位玉米行间杂草,满足基于机器视觉的变量施药系统喷施要求,提出了一种基于行宽的多行玉米行间杂草识别算法.该算法以垂直拍摄的3叶期3行玉米田间图像为研究对象,利用YIQ颜色空间中的Q分量灰度化田间彩色图像,以降低自然光源对图像的影响;通过建立实际田间玉米行宽与图像玉米行宽的映射关系,将3叶期玉米行的宽度映射到对应图像中,并确定基于识别率和运算速度的覆盖范围;以具有一定宽度的玉米行作为识别基准,减小未连通叶片区域的误识别率,提高对杂草识别的精度.从识别精度和速度2方面与基于作物行中心线识别算法进行了对比.研究结果表明,对于3叶期3行玉米田间图像,杂草正确识别率可达89.2％,速度为197 ms.本算法有效地提高了行间杂草识别的精度和速度,能够初步满足基于机器视觉的变量施药系统对大田玉米多行喷施的工作要求.     

基于热红外和RGB图像的番茄苗期高温胁迫检测方法

【目的】针对实际生产场景中番茄苗期生长遇到的高温胁迫问题，提出一种基于热红外和RGB图像的番茄苗期高温胁迫检测方法。【方法】首先，通过番茄苗期热红外图像反演获取番茄冠层温度参数，采用偏最小二乘(Partial least squares, PLS)模型提取冠层温度特征指标；然后，建立采用3种不同主干特征提取网络的MaskRCNN模型，通过迁移学习的方式将番茄苗期RGB图像输入Mask-RCNN模型，进行高温胁迫症状实例分割，得到番茄苗期胁迫症状特征指标；最后，利用提取的温度和胁迫症状特征指标构建分级数据集，输入高温胁迫分级模型，得到高温胁迫等级。【结果】基于PLS模型提取的冠层温度特征指标累计贡献率达95.45%；基于ResNet101+Mask-RCNN的高温胁迫症状分割网络对番茄苗期轻度和重度胁迫的分割精度最高，均值平均查准率(Mean average precision, mAP)分别为77.3%和73.8%；基于温度和胁迫症状特征指标构建的4种高温胁迫分级模型中，反向传播神经网络(Back propagation neural network, BPNN)获得最好的高温胁迫分级...     

基于分级的福建将乐地区栲树树高曲线模型研究

【目的】研究栲树树高曲线模型，为编制福建省将乐地区亚热带常绿阔叶林森林经营数表、指导当地森林经营提供依据。【方法】以将乐地区常绿阔叶林中的栲树为研究对象，以17个栲树林样地中427株栲树胸径、树高实测数据为建模数据，通过分析坡度、坡向和坡位3个地形因子与树高生长之间的相关关系，以Richards方程为基本方程，分别构建基于树高分级和地形分级的栲树树高曲线模型，并与经典树高曲线模型进行精度对比。【结果】基于树高分级的栲树树高曲线模型的决定系数（R²）最大，为0.630;均方根误差（RMSE）最小，为2.283，表明模型拟合精度较高。基于地形分级的栲树树高曲线模型的平均绝对误差（MAE）和平均相对误差（MRE）分别为1.707和0.163，均为最小，表明该模型预估精度最高。【结论】结合将乐地区实际情况，基于地形分级的树高曲线模型能较为精确地预估栲树树高。     

基于改进YOLOX-Nano算法的柑橘梢期长势智能识别

【目的】采用机器视觉技术开展柑橘梢期的智能感知技术研究，以解决背景与目标颜色相似造成识别精度低的问题，实现柑橘梢期自动监测，探索算法的改进方法。【方法】根据不同卷积层提取特征的特点与不同注意力机制的作用，提出了一种基于多注意力机制改进的YOLOX-Nano智能识别模型，建立多元化果园数据集并进行预训练。【结果】改进的YOLOX-Nano算法使用果园数据集作为预训练数据集后，各类别平均精度的平均值(Mean average precision, mAP)达到88.07%。与YOLOV4-Lite系列模型相比，本文提出的改进模型在使用较少的参数和计算量的情况下，识别精度有显著的提升，mAP分别比YOLOV4-MobileNetV3和YOLOV4-GhostNet提升6.58%和6.03%。【结论】改进后的模型在果园监测终端的轻量化部署方面更具有优势，为农情实时感知和智能监测提供了可行的数据和技术解决方案。     

基于迁移学习的棉花识别

提高智能采棉机效率的一个重要途径是实现单个、重叠和遮挡棉花的识别,避免误采摘和漏采摘。针对不同形态棉花的识别,常规的特征提取方法难以达到令人满意的结果,因而采用基于迁移学习的棉花识别方法和基于迁移模型的特征提取与极限学习机(extreme learning machine,ELM)相结合的方法进行棉花识别研究。首先更改AlexNet、GoogleNet、ResNet-50模型分类层和设置相关参数,用训练好的迁移模型对棉花验证集识别,然后利用训练好的迁移模型进行棉花数据集特征提取,再用训练集的特征训练ELM模型,统计不同隐含层神经元个数的ELM模型对棉花的识别准确率。AlexNet、GoogleNet、ResNet-50迁移模型识别率依次为92.03%、93.19%、93.68%;使用特征提取再与ELM结合的方法,准确率比对应迁移模型分别提高了1.97、1.34、1.55百分点。结果表明,迁移模型对小样本棉花识别也有较高准确率,基于特征提取与ELM相结合的方法可进一步提高准确率。     

基于红外热成像技术的玉米种子活力等级分类方法研究

针对传统玉米种子活力等级分类方法耗时长、环境要求严格、对种子产生损伤等问题,利用红外热成像技术结合SVM算法,建立了快速、无损、高效的玉米种子活力等级分类方法。首先采用人工老化的方法将1 200粒玉米种子分组分别老化0 h,72 h,144 h。利用不同老化时间玉米种子具有不同的生理特性,通过红外热成像仪采集温度胁迫后自然冷却的玉米种子红外热像图,提取温度值作为特征。随后对玉米种子进行标准萌发实验,根据实验结果,将玉米种子分为高活力,中活力和低活力3个活力等级。将温度值作为特征,活力等级作为标签分别建立K最近邻(KNN)和支持向量机(SVM)模型并进行训练,以模型分类准确率和训练时间作为评价指标,确定较佳模型,最终通过网格搜索对选择的模型参数进行优化。结果表明基于红外热成像技术结合支持向量机(SVM)建立的模型,训练集准确率达到了92.4%,测试集准确率为91%,训练用时0.12s。该模型经过优化后训练集准确率达到了97.1%,测试集准确率达到了96.5%。     

基于机器视觉的茶叶等级区分

提出利用机器视觉和matlab图像处理技术来区分茶叶的等级。以4个等级的绿茶为实验对象,通过提取不同等级茶叶的图像形状特征参数,采用多类逐步分析法进行特征优化并建立区分模型,实现了室内条件下茶叶等级的区分,正确率达81.25%。     

AdaBoost classifiers for pecan defect classification

One of the constraints in the adoption of machine vision inspection systems for food products is low classification accuracy. This study attempts to improve pecan defect classification accuracy by using machine learning classifiers: AdaBoost and support vector machine (SVM). X-ray images of good and defective pecans, 100 each, were segmented and features were extracted. Twenty classification runs were made to adjust parameters and 300 classification runs to compare classifiers. The Real AdaBoost classifier gave average classification accuracy of 92.2% for the Reverse water flow segmentation method and 92.3% for the Twice Otsu segmentation method. The Linear SVM classifier gave average classification accuracy of 90.1% for the Reverse water flow method and 92.7% for the Twice Otsu method. Computational time for the classifiers varied by two orders of magnitude: Bayesian (10⁻⁴ s), SVM (10⁻⁵ s), and AdaBoost (10⁻⁶ s). AdaBoost classifiers improved classification accuracy by 7% when Bayesian accuracy was poor (less than 89%). The AdaBoost classifiers also adapted well to data variability and segmentation methods. A minimalist AdaBoost classifier, more suitable for real time applications, using fewer features can be built. Overall, the selected AdaBoost classifiers improved classification accuracy, reduced classification time, and performed consistently better for pecan defect classification.     

Identification of nine Iranian wheat seed varieties by textural analysis with image processing

Applying machine vision techniques to classify wheat seeds based on their varieties is an objective method which can increase the accuracy of this process in real applications. In this study, several textural feature groups of seeds images were examined to evaluate their efficacy in identification of nine common Iranian wheat seed varieties. On the whole, 1080 gray scale images of bulk wheat seeds (120 images of each variety) were acquired at a stable illumination condition (florescent ring light). Totally, 131 textural features were extracted from gray level, GLCM (gray level cooccurrence matrix), GLRM (gray level run-length matrix), LBP (local binary patterns), LSP (local similarity patterns) and LSN (local similarity numbers) matrices. The so-called stepwise discrimination method was employed to select and rank the most significant textural features of each matrix individually as well as features of all matrices simultaneously. LDA (linear discriminate analysis) classifier was employed for classification using top selected features. The average classification accuracy of 98.15% was obtained when top 50 of all selected features were used in the classifier. The results confirmed that LSP, LSN and LBP features had a significant influence on the improvement of classification accuracy compared to previous studies.     

转基因抗虫彩色棉产量性状和纤维品质选择的研究

通过对116份转基因抗虫彩色棉两年的间比试验,对产量性状和纤维品质性状进行测定。产量性状单株结铃数、衣分、子指、霜前花、铃重变异系数分别为72.82、75.95、7.27、102.31、5.64。纤维品质指标2.5%跨长、整齐度、伸长率、比强度、马克隆值变异系数分别为17.96、2.12、15.60、34.10、12.88。结果表明彩色棉育种性状选择先后顺序:霜前花>衣分>单株结铃数>比强度>2.5%跨长>伸长率>马克隆值>子指>铃重>整齐度。     

A new approach to aflatoxin detection in chili pepper by machine vision

Aflatoxins are the toxic metabolites of Aspergillus molds, especially by Aspergillus flavus and Aspergillus parasiticus. They have been studied extensively because of being associated with various chronic and acute diseases especially immunosuppression and cancer. Aflatoxin occurrence is influenced by certain environmental conditions such as drought seasons and agronomic practices. Chili pepper may also be contaminated by aflatoxins during harvesting, production and storage. Aflatoxin detection based on chemical methods is fairly accurate. However, they are time consuming, expensive and destructive. We use hyperspectral imaging as an alternative for detection of such contaminants in a rapid and nondestructive manner. In order to classify aflatoxin contaminated chili peppers from uncontaminated ones, a compact machine vision system based on hyperspectral imaging and machine learning is proposed. In this study, both UV and Halogen excitations are used. Energy values of individual spectral bands and also difference images of consecutive spectral bands were utilized as feature vectors. Another set of features were extracted from those features by applying quantization on the histogram of the images. Significant features were selected based on proposed method of hierarchical bottleneck backward elimination (HBBE), Guyon’s SVM-RFE, classical Fisher discrimination power and Principal Component Analysis (PCA). Multi layer perceptrons (MLPs) and linear discriminant analysis (LDA) were used as the classifiers. It was observed that with the proposed features and selection methods, robust and higher classification performance was achieved with fewer numbers of spectral bands enabling the design of simpler machine vision systems.     

Classification of foreign fibers in cotton lint using machine vision and multi-class support vector machine

Automatic classification of foreign fibers in cotton lint using machine vision is still a challenge due to various colors and shapes of the foreign fibers. This paper presents a novel classification method based on multi-class support vector machine (MSVM) which aims at accurate and fast classification of the foreign fibers. Firstly, live images were acquired by a machine vision system and then processed using image processing algorithms. Then the color features, shape features and texture features of each foreign fiber object were extracted and feature vectors were composed. Afterwards, three kinds of multi-class support vector machines were constructed, i.e., one-against-all decision-tree based MSVM, one-against-one voting based MSVM and one-against-one directed acyclic graph MSVM separately. At last, with the extracted feature vectors as input, the MSVMs were tested using leave-one-out cross validation. The results indicate that both the one-against-one voting based MSVM and the one-against-one directed acyclic graph MSVM can satisfy the accuracy requirement of the classification of foreign fibers, and the mean accuracy is 93.57% and 92.34% separately. The one-against-all decision-tree based MSVM only obtains mean accuracy of 79.25% which can not meet the accuracy requirement. In classification speed, one-against-one directed acyclic graph MSVM is the fastest and fitter for online classification.     

低酚棉的丰产育种目标及方法

本文综述了低酚棉育种的国内外研究动态及其进展,着重讨论了有关低酚棉产量及产量构成因素的育种问題。资料表明,随着遗传背景的不断改良,低酚棉与高酚棉的产量差异正在逐渐缩小,许多品种都已赶上甚至超过推广品种。本文提出了在主攻低酚棉产量时,应着重注意提高衣分,并注意单株铃数、铃重、烂铃率和单株结铃性的改良。提高衣分,应在确保较大籽指的基础上,通过提高衣指来实现提高衣分育种目标。另外,本文还对低酚棉的几种育种方法进行了讨论。     

基于多特征融合的茶叶鲜叶等级识别的方法研究

茶叶鲜叶等级直接影响优质绿茶成品的等级,如果在鲜叶阶段就茶叶的芽叶数量进行等级识别,并将不同等级鲜叶分离出来,制作不同等级的绿茶成品,从一定程度上解决了优质绿茶鲜叶采摘环节的难题.提出基于茶叶形态、纹理和HOG特征的鲜叶分级方法,采集鲜叶样本图片,对样本图片进行预处理操作,再提取鲜叶形态和纹理特征等特征参数,建立机器学习模型支持向量机、随机森林和线性判别法K-最近邻对新鲜茶叶样本进行分类,得到各等级的茶叶识别结果.试验结果表明,单独使用一种特征分类效果不佳,也不符合茶叶本身的复杂性.将多种特征融合有更好的分类效果;3种算法中,随机森林算法有较高的优越性,准确率达97.06％.该研究提取的多特征参数和分类模型,为实际鲜叶的生产加工等级识别提供参考.