Evaluating high resolution SPOT 5 satellite imagery for crop identification

High resolution satellite imagery offers new opportunities for crop monitoring and assessment. A SPOT 5 image acquired in May 2006 with four spectral bands (green, red, near-infrared, and short-wave infrared) and 10-m pixel size covering intensively cropped areas in south Texas was evaluated for crop identification. Two images with pixel sizes of 20 m and 30 m were also generated from the original image to simulate coarser resolution satellite imagery. Two subset images covering a variety of crops with different growth stages were extracted from the satellite image and five supervised classification techniques, including minimum distance, Mahalanobis distance, maximum likelihood, spectral angle mapper (SAM), and support vector machine (SVM), were applied to the 10-m subset images and the two coarser resolution images to identify crop types. The effects of the short-wave infrared band and pixel size on classification results were also examined. Kappa analysis showed that maximum likelihood and SVM performed better than the other three classifiers, though there were no statistical differences between the two best classifiers. Accuracy assessment showed that the 10-m, four-band images based on maximum likelihood resulted in the best overall accuracy values of 91% and 87% for the two respective sites. The inclusion of the short-wave infrared band statistically significantly increased the overall accuracy from 82% to 91% for site 1 and from 75% to 87% for site 2. The increase in pixel size from 10 m to 20 m or 30 m did not significantly affect the classification accuracy for crop identification. These results indicate that SPOT 5 multispectral imagery in conjunction with maximum likelihood and SVM classification techniques can be used for identifying crop types and estimating crop areas.     

基于无人机低空遥感的水稻田间杂草分布图研究

目的 获取水稻田的低空遥感图像并分析得到杂草分布图，为田间杂草精准施药提供参考。方法 使用支持向量机(SVM)、K最近邻算法(KNN)和AdaBoost 3种机器学习算法，对经过颜色特征提取和主成分分析(PCA)降维后的无人机拍摄的水稻田杂草可见光图像进行分类比较；引入一种无需提取特征和降维、可自动获取图像特征的卷积神经网络(CNN)，对水稻田杂草图像进行分类以提升分类精度。结果 SVM、KNN和AdaBoost对测试集的测试运行时间分别为0.500 4、2.209 2和0.411 1 s，分类精度分别达到89.75%、85.58%和90.25%，CNN对图像的分类精度达到92.41%，高于上述3种机器学习算法的分类精度。机器学习算法及CNN均能有效识别水稻和杂草，获取杂草的分布信息，生成水稻田间的杂草分布图。结论 CNN对水稻田杂草的分类精度最高，生成的水稻田杂草分布图效果最好。     

基于机器视觉的脐橙品质在线分级检测系统

针对脐橙自动分级检测中存在正确识别率偏低、实时性不强的问题,提出一种综合特征提取方法:在对图像颜色模型进行转换后,用H分量图像提取脐橙的大小特征;S分量图像通过背景分割、边缘灰度补偿、整体亮度变换后提取脐橙的果面缺陷特征;采用R、G、R-G3个分量的均值和标准差提取脐橙的颜色特征。以脐橙的大小特征、果面缺陷特征和颜色特征为支持向量机(Support vector machine,SVM)的试验输入向量,进行脐橙分级检测试验,以实现提高脐橙自动分级正确识别率和增强实时。试验结果表明:该SVM分类器对脐橙分级的正确识别率为91.5%,处理时间为160ms,适合于实时环境下的分级检测。     

基于不同分类器的农用地分类提取

【目的】分析Landsat 8 OLI卫星遥感影像数据面向农用地分类的实际应用方法和效果,以新疆奇台县南部为研究对象。【方法】使用随机森林(RF)、支持向量机(SVM)和神经网络(Neural Net)三种分类器进行研究区农用地分类对比。【结果】通过对三种分类器参数设置参数精度检验,利用上述三种算法对农用地地物分类进行精度评价,在整体分类精度中,支持向量机算法(SVM)<随机森林算法(RF)<神经网络算法(Neural Net),分类精度分别为:90.75%,94.30%和94.84%。【结论】神经网络方法(Neural Net)在该地区的农用地物整体分类上,比支持向量机(SVM)和随机森林法(RF)相比具有一定的优势,并获得较好的分类精度。     

基于近红外光谱的纽荷尔脐橙产地识别研究

【目的】中国柑橘产区分布广、生态类型复杂,不同产地纽荷尔脐橙果实品质和市场效应具有较大差异。研究基于近红外光谱技术的柑橘产地识别技术,利于不同柑橘产地果品的识别和鉴伪。【方法】从中国南方17个纽荷尔脐橙主要产地选择代表性成年果园,分别采摘成熟鲜果样品100个。利用SupNIR-1500近红外分析仪采集脐橙果实赤道部、肩部表面以及果汁滤液的近红外反射光谱,光谱波长范围为1 000-2 499 nm。采用主成分分析法对原始光谱数据进行预处理,提取近红外光谱的特征信息以降低数据集维度以及噪声。研究人工神经网络理论,构建由一个输入层、一个具有非线性激励函数的隐藏层和一个输出层组成的典型的3层人工神经网络识别模型。研究由径向基函数作为核函数、以光谱主成分作为输入的支持向量机模型,构建由126个分类器组成的一对一扩展支持向量机模型。研究遗传算法优异的自然选择特性,利用遗传算法从光谱主成分中选择出最优的特征基因子集作为支持向量机的输入,构建遗传算法-支持向量机模型。利用3种模型分别对果汁滤液的近红外反射光谱数据进行分类,从而实现产地识别测试,并根据产地识别精度筛选出最优的产地识别模型。进一步对比该最优识别模型对果实赤道部、肩部反射光谱数据的识别精度,从而确定识别精度最高的光谱数据采集源。【结果】利用所建立的3层人工神经网络模型对纽荷尔脐橙果汁滤液的近红外光谱进行产地识别测试,确定当输入神经元数量为11、隐藏神经元数量为13时,模型对果实产地识别的最佳精度达81.45%。采用一对一扩展方式建立支持向量机产地识别模型,研究确定采用径向基函数作为核函数,当主成分数量为20时,脐橙产地识别精度最高可达86.98%。测试利用遗传算法-支持向量机混合模型进行脐橙产地分类识别,确定当种群数量为200、遗传代数为100、交叉概率0.7、突变概率0.01时,遗传算法选择出最优的基因子集进行产地识别,遗传算法-支持向量机模型的产地识别精度最高可达89.72%,优于人工神经网络分类模型和支持向量机分类模型的产地识别精度。进一步利用遗传算法-支持向量机产地识别模型对果实赤道部及肩部的果面反射光谱进行产地识别测试,得到对应的最高识别精度分别为80.00%和69.00%。【结论】遗传算法-支持向量机模型对果汁反射近红外光谱进行产地识别精度最高,优于人工神经网络模型和支持向量机模型。该模型对果实赤道部反射光谱进行分类的精度次于果汁滤液反射光谱但优于果实肩部反射光谱,因此,可利用赤道部的反射光谱实现非破坏性果实产地分类识别。     

基于EnMAP-Box的遥感图像分类研究

采用2007年6月云南省勐腊县TM遥感数据,利用EnMAP-box进行了支持向量机的图像分类研究,以网格搜索法寻找最优参数,在设定的范围内,求得了最优C和g参数,用此参数进行支持向量机的遥感图像土地覆盖分类。结果表明：SVM方法较最大似然分类方法具有较高的分类精度,特别是阔叶林和橡胶林的精度明显优于最大似然分类方法;对于面积较小的次要类型,2种分类方法的精度基本保持一致;SVM的总体精度相对于最大似然分类提高了11.9％。     

Support vector machine approach to real-time inspection of biscuits on moving conveyor belt

An intelligent system for colour inspection of biscuit products is proposed. In this system, the state-of-the-art classification techniques based on Support Vector Machines (SVM) and Wilk's λ analysis were used to classify biscuits into one of four distinct groups: under-baked, moderately baked, over-baked, and substantially over-baked. The accuracy of the system was compared with standard discriminant analysis using both direct and multi-step classifications. It was discovered that the radial basis SVM after Wilk's λ was more precise in classification compared to other classifiers. Real-time implementation was achieved by means of multi-core processor with advanced multiple-buffering and multithreading algorithms. The system resulted in correct classification rate of more than 96% for stationary and moving biscuits at 9 m/min. It was discovered that touching and non-touching biscuits did not significantly interfere with accurate assessment of baking. However, image processing of touching biscuits was considerably slower compared to non-touching biscuits, averaging at 36.3 ms and 9.0 ms, respectively. The decrease in speed was due to the complexity of the watershed-based algorithm used to segment touching biscuits. This image computing platform can potentially support the requirements of the high-volume biscuit production.     

Application of distributed SVM architectures in classifying forest data cover types

In many ‘real-world’ applications, a classification of large data sets, which are often also imbalanced, is difficult due to the small, but usually more interesting classes. In this study, a large data set, forest cover type classes, which is actually multi-class classification defined with seven imbalanced classes and used as a resource inventory information was analyzed and evaluated. The data set was transformed into seven new data sets and a support vector machine (SVM) was employed to solve a binary classification problem of balanced and imbalanced data sets with various sizes. In the two approaches considered, the use of distributed SVM architectures, which basically reduces the complexity of the quadratic optimization problem of very large data sets, and the use of two sampling approaches for classification of imbalanced data sets were combined and results presented. The experimental results of distributed SVM architectures show the improvement of the accuracy for larger data sets in comparison to a single SVM classifier and their ability to improve the correct classification of the minority class.     

Comparison of machine learning algorithms for mapping mango plantations based on Gaofen-1 imagery

Mango is a commercial crop on Hainan Island, China, that is cultivated to develop the tropical rural economy. The development of accurate and up-to-date maps of the spatial distribution of mango plantations is necessary for agricultural monitoring and decision management by the local government. Pixel-based and object-oriented image analysis methods for mapping mango plantations were compared using two machine learning algorithms (support vector machine (SVM) and Random Forest (RF)) based on Chinese high-resolution Gaofen-1 (GF-1) imagery in parts of Hainan Island. To assess the importance of different features on classification accuracy, a combined layer of four original bands, 32 gray-level co-occurrence (GLCM) texture indices, and 10 vegetation indices were used as input features. Then five different sets of variables (5, 10, 20, and 30 input variables and all 46 variables) were classified with the two machine learning algorithms at object-based level. Results of the feature optimization suggested that homogeneity and variance were very important variables for distinguishing mango plantations patches. The object-based classifiers could significantly improve overall accuracy between 2–7% when compared to pixel-based classifiers. When there were 5 and 10 input variables, SVM showed higher classification accuracy than RF, and when the input variables exceeded 20, RF showed better performances. After the accuracy achieved saturation points, there were only slightly classification accuracy improvements along with the numbers of feature increases for both of SVM and RF classifiers. The results indicated that GF-1 imagery can be successfully applied to mango plantation mapping in tropical regions, which would provide a useful framework for accurate tropical agriculture land management.     

基于面向对象的资源3号遥感影像森林分类研究

森林类型识别技术是遥感分类中的重点和难点,采用面向对象的遥感影像分类方法是实现森林类型分类的新方法。资源3号遥感影像可为森林类型提取提供新方向。以资源3号遥感影像作为基础研究数据,采用面向对象的分类方法,选择分形网络演化法进行多尺度分层分割,并结合典型地物的光谱特征、纹理特征、几何特征以及植被指数,构建了适用于森林类型提取的决策树模型,并与分割尺度不同的支持向量机分类方法进行比较分析。结果表明:多层分割的决策树分类方法分类精度高于单层分割的支持向量机分类方法,分类精度分别提高了6.1%和12.5%。说明建立多层分割的决策树分类方法适用于森林类型的分类研究。     

基于图像识别的无人机精准喷雾控制系统的研究

【目的】针对传统的植保无人机喷雾作业时化肥农药浪费大,利用率低,造成环境污染的问题,研制一种基于图像识别的无人机精准喷雾控制系统。【方法】利用中值滤波算法对田间航拍图像进行去噪,采用分层K_means硬聚类算法实现对农田航拍图像的分割,提取非作物区域的颜色、纹理特征空间的22个特征参数,设计支持向量机分类器进行分类识别。根据优选的17个特征参数,利用以径向基函数作为核函数的支持向量机对非作物区域图像进行识别,并根据识别结果控制喷头,实现精准喷雾。【结果】测试样本的识别率可达为76.56%,在无干扰风场情况下,当P_阀为10%时,减施率可达32.7%。【结论】本系统为农业航空精准喷雾控制技术的应用提供了参考方向和决策支持。     

基于Android手机的树木叶片识别系统

为了准确高效地识别树木叶片,开发了一款基于Android操作平台的树木叶片设别系统.该系统提取13种树木叶片特征描述子,选择支持向量机作为分类器.该系统包括图像获取、图像处理、特征提取、分类识别和结果展示5个模块.选取来自15个树种的1 500片树叶进行了试验,结果表明,该系统的平均识别率可以达到94.44％,优于BP神经网络的91.56％,达到了令人满意的效果.该系统特征描述子的筛选、提取以及分类器算法还可以进一步优化,以更好地满足用户需求.     

基于Android手机的树木叶片识别系统

为了准确高效地识别树木叶片,开发了一款基于Android 操作平台的树木叶片设别系统。该系统提取13 种树木叶片特征描述子,选择支持向量机作为分类器。该系统包括图像获取、图像处理、特征提取、分类识别和结果展示5 个模块。选取来自15 个树种的1 500 片树叶进行了试验,结果表明,该系统的平均识别率可以达到94.44%,优于BP 神经网络的91.56%,达到了令人满意的效果。该系统特征描述子的筛选、提取以及分类器算法还可以进一步优化,以更好地满足用户需求。     

基于混沌机制的人工蜂群算法优化的支持向量机分类器

支持向量机的分类性能在很大程度上取决于其相关参数的选择,为了改善支持向量机的分类准确率,本文采用基于混沌机制的人工蜂群算法对其参数进行优化。在传统人工蜂群算法的基础上,采用Logistic混沌映射初始化种群和锦标赛选择策略,进一步提高人工蜂群算法的收敛速度和寻优精度。该方法采用分类准确率作为适应度函数,利用人工蜂群算法对支持向量机的惩罚因子和核函数参数进行优化。通过对多个标准数据集的分类测试,证明基于混沌机制的人工蜂群算法优化的支持向量机分类器能够获得更高的分类准确率。     

一种基于支持向量机的图像检索方法

提出了一种基于支持向量机(Support Vector Machine,SVM)的个性图像检索方法,首先融合符合用户需求的图像的物理特征构造SVM分类器,然后把获得的图像信息提交给分类器进行识别,最后把检索结果返回给用户.实验结果表明,用SVM作为学习机器可以实现对图像的检索分类.     

基于特征融合的随机森林模型茶鲜叶分类

目的 解决机采茶鲜叶中混有不同等级的茶叶,且混杂度高、物理特征分类精确度低的问题。方法 利用随机森林分类模型,提出一种基于颜色和边缘特征融合的方法。试验采集3种不同等级的茶鲜叶,对原始图像进行裁剪、尺寸归一化和去噪等处理,再进行颜色特征和边缘特征提取。通过参数的修改和测试,构建最优的随机森林分类模型,并且同K最近邻、SVM分类器进行对比试验。结果 特征融合之后随机森林模型的分类准确率达到99.45%,比单一颜色特征和边缘特征的分类准确率分别高7.14和9.34个百分点;比K最近邻和SVM分类器准确率分别高15.38和5.49个百分点。结论 所建立的方法能够对茶鲜叶单芽、一芽一叶、一芽二叶进行精确的分类。     

Hyperspectral species mapping for automatic weed control in tomato under thermal environmental stress

This work studied the impacts of variations in environmental temperature on hyperspectral imaging features in the visible and near infrared regions for robust species identification for weed mapping in tomato production. Six major Californian processing tomato cultivars, black nightshade (Solanum nigrum L.) and redroot pigweed (Amaranthus retroflexus L.) were grown under a variety of diurnal temperature ranges simulating conditions common in the Californian springtime planting period and one additional treatment simulating greenhouse growing conditions. The principal change in canopy reflectance with varying temperature occurred in the 480-670 and 720-810 nm regions. The overall classification rate ranged from 62.5% to 91.6% when classifiers trained under single temperatures were applied to plants grown at different temperatures. Eliminating the 480-670 nm region from the classifier’s feature set mitigated the temperature effect by stabilizing the total crop vs. weed classification rate at 86.4% over the temperature ranges. A site-specific recalibration method was also successful in alleviating the bias created by calibrating the models on the extreme temperatures and increased the classification accuracy to 90.3%. A global calibration method, incorporating all four temperature conditions in the classifier feature space, provided the best average total classification accuracy of 92.2% out of the methods studied, and was fairly robust to the varying diurnal temperature conditions.     

Using an image segmentation and support vector machine method for identifying two locust species and instars

Locusts are agricultural pests around the world. To cognize how locust distribution density and community structure are related to the hydrothermal and vegetation growth conditions of their habitats and thereby providing rapid and accurate warning of locust invasions, it is important to develop efficient and accurate techniques for acquiring locust information. In this paper, by analyzing the differences between the morphological features of Locusta migratoria manilensis and Oedaleus decorus asiaticus, we proposed a semi-automatic locust species and instar information detection model based on locust image segmentation, locust feature variable extraction and support vector machine(SVM) classification. And we subsequently examined its applicability and accuracy based on sample image data acquired in the field. Locust image segmentation experiment showed that the proposed GrabCut-based interactive segmentation method can be used to rapidly extract images of various locust body parts and exhibits excellent operability. In a locust feature variable extraction experiment, the textural, color and morphological features of various locust body parts were calculated. Based on the results, eight feature variables were selected to identify locust species and instars using outlier detection, variable function calculation and principal component analysis. An SVM-based locust classification experiment achieved a semi-automatic detection accuracy of 96.16% when a polynomial kernel function with a penalty factor parameter c of 2 040 and a gamma parameter g of 0.5 was used. The proposed detection model exhibits advantages such as high applicability and accuracy when it is used to identify locust instars of L. migratoria manilensis and O. decorus asiaticus, and it can also be used to identify other species of locusts.     

基于颜色和纹理特征的大麦主要病害识别研究

为提出一种在自然环境条件下基于采集图像的颜色一阶与二阶矩和纹理LBP算子改进模式综合特征参数的大麦病害识别方法,以甘肃河西地区发生的大麦白粉病、云纹病和条锈病为研究对象,采用颜色矩和LBP算子均匀模式综合特征参数来提取大麦病斑的颜色和纹理特征,并将该特征向量作为输入向量构建以径向基为核函数的支持向量机(SVM)分类器模型。利用SVM分类模型对采集到的355幅病害图像进行实例分析,结果表明当径向基参数时,大麦病害整体识别正确率达84.7458%。本研究为农田大麦病害诊断提供了有效的分析手段,验证分类模型在大麦病害研究中的可行性,并可为其他农作物病害诊断提供借鉴和参考。     

Aerodynamic analysis and CFD simulation of several cellulose evaporative cooling pads used in Mediterranean greenhouses

The present work makes an aerodynamic analysis and computational fluid dynamics (CFD) simulation of the four commercial models of corrugated cellulose evaporative cooling pads that are most widely used in Mediterranean greenhouses. The geometric characteristics of the pads have been determined as well as the volume of water they retain at different flows of water, thus obtaining the mean thickness of the sheet of water which runs down them and their porosity. By means of low velocity wind tunnel experiments, the pressure drop produced by the pads has been recorded at different wind speeds and water flows. In this way it has been possible to obtain the relationship of the permeability and the inertial factor with pad porosity using a cubic type equation. Finally, a CFD simulation with a 3D model has been carried out for both dry pads (Q_w = 0 l s⁻¹ m⁻²) and wet ones (Q_w = 0.256 l s⁻¹ m⁻²), finding good correlation between the simulated and experimental pressure drop, with maximum differences of 9.08% for dry pads and 15.53% for wet ones at an airspeed of 3 m s⁻¹.