The recognition of apple fruits in plastic bags based on block classification

The recognition of apple fruits in plastic bags is easy to be affected by reflected and refracted light. In order to weaken the influence of light, a method based on block classification is proposed. The method adopts watershed algorithm to segment original images into irregular blocks based on edge detection results of R–G grayscale images firstly. Compared with the watershed algorithm based on gradient images, the segmentation method can preserve fruits edge and reduce the number of blocks by 20.31%, because graying image method, R–G, filters most of leaves and edge detection operator insures that the edge of fruits are detected accurately. Next, these blocks are classified into fruit blocks and non-fruit blocks by support vector machine on the basis of the color and texture features extracted from blocks. Compared with the image recognition method based on pixel classification, the proposed method can restrain the interference of light caused by plastic bags effectively. The false negative rate (FNR) and false positive rate (FPR) of the method based on pixel classification are 21.71 and 14.53% respectively. The FNR and FPR of the proposed method are 4.65 and 3.50% respectively.     

Comparison of chlorophyll fluorescence curves and texture analysis for automatic plant identification

With automatic plant identification methods, the amount of herbicides used in agriculture can be reduced when herbicides are sprayed only on weeds. In the present study, leaves of oat (Avena sativa) and dandelion (Taraxacum officinale, TAROF) were arranged so that there was overlap between the species, imaged with a pulse amplitude modulation fluorescence camera and photographed with a digital color camera. The fluorescence induction curves from each pixel were parameterized to obtain a set of features and from color photographs, texture features were calculated. A support vector algorithm that also performed feature selection was used for pattern recognition of both data sets. Fluorescence-based identification worked well with oat leaves, producing 92.2 % of correctly identified pixels, whereas the texture-based method often mis-identified the central vein of a TAROF leaf as oat, identifying correctly only 66.5 % of oat pixels. With TAROF that shows a clear dicot-type texture, the texture method was slightly better (96.4 % correctly identified pixels) than the fluorescence method (94.6 %). In fluorescence-based identification, the accuracy varied between entire TAROF leaves, probably reflecting the genetic variability of TAROF. The results suggest that the accuracy of identification could be improved by combining two identification methods.     

Identification of insect-damaged wheat kernels using short-wave near-infrared hyperspectral and digital colour imaging

Healthy wheat kernels and wheat kernels damaged by the feeding of the insects: rice weevil (Sitophilus oryzae), lesser grain borer (Rhyzopertha dominica), rusty grain beetle (Cryptolestes ferrugineus), and red flour beetle (Tribolium castaneum) were scanned using a near-infrared (NIR) hyperspecrtal imaging system (700-1100 nm wavelength range) and a colour imaging system. Dimensionality of hyperspectral data was reduced and statistical and histogram features were extracted from NIR images of significant wavelengths and given as input to three statistical discriminant classifiers (linear, quadratic, and Mahalanobis) and a back propagation neural network (BPNN) classifier. A total of 230 features (colour, textural, and morphological) were extracted from the colour images and the most contributing features were selected and used as input to the statistical and BPNN classifiers. The quadratic discriminant analysis (QDA) classifier gave the highest accuracy and correctly identified 96.4% healthy and 91.0-100.0% insect-damaged wheat kernels using the top 10 features from 230 colour image features combined with hyperspectral image features.     

RGB and multispectral UAV image fusion for Gramineae weed detection in rice fields

In this paper, a new method to fuse low resolution multispectral and high resolution RGB images is introduced, in order to detect Gramineae weed in rice fields with plants at 50 days after emergence (DAE).The images are taken from a fixed-wing unmanned aerial vehicle (UAV) at 60 and 70 m altitude. The proposed method combines the texture information given by a high resolution red–green–blue (RGB) image and the reflectance information given by a low resolution multispectral (MS) image, to obtain a fused RGB-MS image with better weed discrimination features. After analyzing the normalized difference vegetation index (NDVI) and normalized green red difference index (NGRDI) for weed detection, it was found that NGRDI presents better features. The fusion method consists of decomposing the RGB image using the intensity, hue and saturation (IHS) transformation, then, a second order Haar wavelet transformation is applied to the intensity layer (I) and the NGRDI image. From this transformation, the low–low (LL) coefficients of the NGRDI image are replaced by the LL coefficients of the I layer. Finally, the fused image is obtained by transforming the new wavelet coefficients to RGB space. To test the method, a one hectare experimental plot with rice plants at 50 DAE with Gramineae weeds was selected. Additionally, to compare the performance of the method, two indices were used, specifically, the M/M_GT index which is the percentage of detected weed area, and the MP index which indicates the precision of weed detection. These indices were evaluated in four validation zones using three Neural Networks (NN) detection systems based on three types of images; namely, RGB, RGB + NGRDI, and fused RGB-NGRDI. The best weed detection performance was obtained by the NN with the fused image, with M/M_GT index between 80 and 108% and MP between 70 and 85%.     

Machine vision for counting fruit on mango tree canopies

Machine vision technologies hold the promise of enabling rapid and accurate fruit crop yield predictions in the field. The key to fulfilling this promise is accurate segmentation and detection of fruit in images of tree canopies. This paper proposes two new methods for automated counting of fruit in images of mango tree canopies, one using texture-based dense segmentation and one using shape-based fruit detection, and compares the use of these methods relative to existing techniques:—(i) a method based on K-nearest neighbour pixel classification and contour segmentation, and (ii) a method based on super-pixel over-segmentation and classification using support vector machines. The robustness of each algorithm was tested on multiple sets of images of mango trees acquired over a period of 3 years. These image sets were acquired under varying conditions (light and exposure), distance to the tree, average number of fruit on the tree, orchard and season. For images collected under the same conditions as the calibration images, estimated fruit numbers were within 16 % of actual fruit numbers, and the F1 measure of detection performance was above 0.68 for these methods. Results were poorer when models were used for estimating fruit numbers in trees of different canopy shape and when different imaging conditions were used. For fruit-background segmentation, K-nearest neighbour pixel classification based on colour and smoothness or pixel classification based on super-pixel over-segmentation, clustering of dense scale invariant feature transform features into visual words and bag-of-visual-word super-pixel classification using support vector machines was more effective than simple contrast and colour based segmentation. Pixel classification was best followed by fruit detection using an elliptical shape model or blob detection using colour filtering and morphological image processing techniques. Method results were also compared using precision–recall plots. Imaging at night under artificial illumination with careful attention to maintaining constant illumination conditions is highly recommended.     

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IHS变换和Mean Shift算法的草地分类研究

为高效估算草地生物量,需要一种方法来提高草地分类精度和降低数据处理时间。该研究基于原始RGB图像采用IHS变换,进行绿度波段图像的融合,并对融合后的图像进行Mean Shift算法分类。结果表明,1）基于IHS图像的草地分类,在视觉上与实际地物更为吻合;2）与其他文献方法对比,本研究方法性能优越,精度达到95%以上;3）可以批量处理多张图像,提高了数据处理效率。     

Immature green citrus fruit detection and counting based on fast normalized cross correlation (FNCC) using natural outdoor colour images

A fast normalized cross correlation (FNCC) based machine vision algorithm was proposed in this study to develop a method for detecting and counting immature green citrus fruit using outdoor colour images toward the development of an early yield mapping system. As a template matching method, FNCC was used to detect potential fruit areas in the image, which was the very basis for subsequent false positive removal. Multiple features, including colour, shape and texture features, were combined in this algorithm to remove false positives. Circular Hough transform (CHT) was used to detect circles from images after background removal based on colour components. After building disks centred in centroids resulted from both FNCC and CHT, the detection results were merged based on the size and Euclidian distance of the intersection areas of the disks from these two methods. Finally, the number of fruit was determined after false positive removal using texture features. For a validation dataset of 59 images, 84.4 % of the fruits were successfully detected, which indicated the potential of the proposed method toward the development of an early yield mapping system.     

Identification of broad-leaved dock (Rumex obtusifolius L.) on grassland by means of digital image processing

Digital image processing has the potential to support the identification of plant species required for site-specific weed control in grassland swards. The present study focuses on the identification of one of the most invasive and persistent weed species on European grassland, the broad-leaved dock (Rumex obtusifolius L., R.o.), in complex mixtures of perennial ryegrass with R.o. and other herbs.A total of 108 digital photographs were obtained from a field experiment under constant recording geometry and illumination conditions. An object-oriented image classification was performed. Image segmentation was done by transforming the red, green, blue (RGB) colour images to greyscale intensity images. Based on that, local homogeneity images were calculated and a homogeneity threshold (0.97) was applied to derive binary images. Finally, morphological opening was performed. The remaining contiguous regions were considered to be objects. Features describing shape, colour and texture were calculated for each of these objects. A Maximum-likelihood classification was done to discriminate between the weed species. In addition, rank analysis was used to test how combinations of features influenced the classification result.The detection rate of R.o. varied with the training dataset used for classification. Average R.o. detection rates ranged from 71 to 95% for the 108 images, which included more than 3,600 objects. Misclassifications of R.o. occurred mainly with Plantago major (P.m.). Between 9 and 16% R.o. objects were classified incorrectly as P.m. and 17–24% P.m. objects were misclassified as R.o. The classification result was influenced by the defined object classes (R.o., P.m., T.o., soil, residue vs. R.o., residue). For instance, classification rates were 86–91% and 65–82% for R.o. exclusively and R.o. against the remaining herb species, respectively.     

基于纹理模型的檀香咖啡豹蠹蛾虫害图像诊断方法研究

基于纹理特征模型的檀香咖啡豹蠹蛾图像诊断方法,根据健康图像和虫害图像在纹理方面表现出的差异,提出海南省北部县市檀香受咖啡豹蠹蛾虫害“多纹理特征”的确定方法。针对每种图像类型,使用提取出的4维多纹理特征,组合得到6种数学模型,并对其进行评估。结果表明：模型1(自变量为熵值均值-相关性均值,因变量为熵值均值-能量均值)的模型精度与分类精度均为最佳,并且总体分类精度达到91.25%。与逐步聚类算法和K-means聚类算法、Logistic模型二分类法相比,该方法在保证分类精度的前提下减小了计算量,并为之后纹理图像分类提供了参考依据。     

Estimation of rice protein content before harvest using ground-based hyperspectral imaging and region of interest analysis

Protein content, which represents rice taste quality, must be estimated in order to create a harvesting plan as well as next year’s basal dressing fertilizer application plan. Ground-based hyperspectral imaging with high resolution (1 × 1 mm per pixel) was used for estimating the protein content of brown rice before harvest. This paper compares the estimation accuracy of rice protein content estimation models generated from the mean reflectances of five regions of interest (ROIs): the overall target area, dark area (less illuminated parts of the rice plants), canopy area (leaves, yellow leaves, and ears), leaf area, and ear and yellow leaf area. The size of the target sampling area was 0.85 × 0.85 m. An R + G + B histogram and a GNDVI–NDVI image were used to separate the target area into the individual ROIs. The values of the coefficient of determination R ² and the root mean square error of prediction (RMSE) were similar for each model: R ² ranged from 0.83 to 0.86 and RMSE ranged from 0.27 to 0.30% for all models except for the dark area model, where R ² = 0.76 and RMSE = 0.35%. There were no significant differences in the magnitude of the estimation error among all models. This result indicates that it is not necessary to obtain an image with a ground resolution that is greater than 0.85 × 0.85 m per pixel to estimate rice protein content before harvest. This result should provide useful information when deciding the altitude of platforms for imaging rice fields.     

基于无人机多光谱遥感图像的玉米田间杂草识别

【目的】为了精确高效识别玉米田间杂草,减少除草剂施用,提高玉米种植管理精准性。【方法】通过六旋翼无人机搭载多光谱相机获取玉米田块多光谱图像。为分离图像中植被与非植被像元,计算了7种植被指数,采用最大类间方差法提取植被指数图像中非植被区域,制作掩膜文件并对多光谱图像掩膜。通过主成分分析对多光谱图像进行变换,保留信息量最多的前3个主成分波段。将试验区域分为训练区域和验证区域,在训练区域中分别选取了675处玉米和525处杂草样本对监督分类模型进行训练,在验证区域选取了240处玉米样本及160处杂草样本评价模型分类精度。将7种植被指数、3个主成分波段的24个纹理特征及经过滤波的10个反射率,共计41项特征作为样本特征参数。利用支持向量机-特征递归消除算法(support vector machines-feature recursive elimination,SVM-RFE)和Relief算法从41项特征中各筛选14项特征构成特征子集,采用支持向量机、K-最近邻、Cart决策树、随机森林和人工神经网络对特征子集进行监督分类。【结果】支持向量机与随机森林对全部特征及2个特征子集分类效果较好,支...     

Assessing the accuracy of mosaics from unmanned aerial vehicle (UAV) imagery for precision agriculture purposes in wheat

High spatial resolution images taken by unmanned aerial vehicles (UAVs) have been shown to have the potential for monitoring agronomic and environmental variables. However, it is necessary to capture a large number of overlapped images that must be mosaicked together to produce a single and accurate ortho-image (also called an ortho-mosaicked image) representing the entire area of work. Thus, ground control points (GCPs) must be acquired to ensure the accuracy of the mosaicking process. UAV ortho-mosaics are becoming an important tool for early site-specific weed management (ESSWM), as the discrimination of small plants (crop and weeds) at early growth stages is subject to serious limitations using other types of remote platforms with coarse spatial resolutions, such as satellite or conventional aerial platforms. Small changes in flight altitude are crucial for low-altitude image acquisition because these variations can cause important differences in the spatial resolution of the ortho-images. Furthermore, a decrease of flying altitude reduces the area covered by each single overlapped image, which implies an increase of both the sequence of images and the complexity of the image mosaicking procedure to obtain an ortho-image covering the whole study area. This study was carried out in two wheat fields naturally infested by broad-leaved and grass weeds at a very early phenological stage. The geometric accuracy differences and crop line alignment among ortho-mosaics created from UAV image series were investigated while taking into account three different flight altitudes (30, 60 and 100 m) and a number of GCPs (from 11 to 45). The results did not show relevant differences in geo-referencing accuracy on the interval of altitudes studied. Similarly, the increase of the number of GCPs did not imply a relevant increase of geo-referencing accuracy. Therefore, the most important parameter to consider when choosing the flying altitude is the ortho-image spatial resolution required rather than the geo-referencing accuracy. Regarding the crop mis-alignment, the results showed that the overall errors were less than twice the spatial resolution, which did not break the crop line continuity at the studied spatial resolutions (pixels from 7.4 to 24.7 mm for 30, 60 and 100 m flying altitudes respectively) on the studied crop (early wheat). The results lead to the conclusion that a UAV flying at a range of 30 to 100 m altitude and using a moderate number of GCPs is able to generate ultra-high spatial resolution ortho-imagesortho-images with the geo-referencing accuracy required to map small weeds in wheat at a very early phenological stage. This is an ambitious agronomic objective that is being studied in a wide research program whose global aim is to create broad-leaved and grass weed maps in wheat crops for an effective ESSWM.     

基于图像处理技术的四种苜蓿叶部病害的识别

基于图像处理技术,对4种苜蓿叶部病害进行识别研究。利用结合K中值聚类算法和线性判别分析的分割方法对病斑图像作分割,获得了较好的分割效果。结果表明:该分割方法在由4种病害图像数据集整合成的汇总图像数据集上综合得分的平均值和中值分别为0.877 1和0.899 7;召回率的平均值和中值分别为0.829 4和0.851 4;准确率的平均值和中值分别为0.924 9和0.942 4。进一步提取病斑图像的颜色特征、形状特征和纹理特征共计129个,利用朴素贝叶斯方法和线性判别分析方法建立病害识别模型,并结合顺序前向选择方法实现特征筛选,分别获得最优特征子集;同时利用这2个最优特征子集,结合支持向量机(Support vector machine,SVM)建立病害识别模型。比较各模型的识别效果,发现利用所建线性判别分析模型下的最优特征子集,结合SVM建立的病害识别模型识别效果最好,训练集识别正确率为96.18%,测试集识别正确率为93.10%。由此可见,本研究所建基于图像处理技术的病害识别模型可用于识别上述4种苜蓿叶部病害,为苜蓿病害的诊断和鉴别提供了一定依据。     

基于国产GF-2遥感影像的大麻地块提取方法研究 ——以安徽省六安市苏埠镇为例

针对传统的基于像元的分类方法提取大麻地块结果存在较为破碎、精度较低的问题,以国产"高分二号"(GF-2)4m的多光谱遥感影像为数据源,在安徽省六安市苏埠镇选取了一个研究区,使用基于规则集的面向对象的方法实现了大麻地块的精确提取。首先,对研究区预处理过的GF-2遥感影像进行多尺度分割,在多尺度分割结果的基础上,确定提取大麻地块的最优分割尺度。其次,针对不同地物类型选取样本对象生成光谱曲线,分析大麻地块与其他地物类型的异同点,并基于光谱分析结果构建规则集最终实现大麻地块的提取。最后,将基于规则集的面向对象分类结果和基于像元分类(监督分类)的结果进行对比分析。结果表明,基于规则集的面向对象方法可以有效的提取出研究区内的大麻地块,精度可以达到91.09%,解决了传统基于像元分类方法提取大麻地块结果较为破碎的问题。     

Mapping crop water stress index in a ‘Pinot-noir’ vineyard: comparing ground measurements with thermal remote sensing imagery from an unmanned aerial vehicle

Characterizing the spatial variability in water status across vineyards is a prerequisite for precision irrigation. The crop water stress index (CWSI) indicator was used to map the spatial variability in water deficits across an 11-ha ‘Pinot noir’ vineyard. CWSI was determined based on canopy temperatures measured with infrared temperature sensors placed on top of well-watered and water-stressed grapevines in 2009 and 2010. CWSI was correlated with leaf water potential (Ψ_L) (R ² = 0.83). This correlation was also tested with results from high resolution airborne thermal imagery. An unmanned aerial vehicle equipped with a thermal camera was flown over the vineyard at 07:30, 09:30, and 12:30 h (solar time) on 31 July 2009. At about the same time, Ψ_L was measured in 184 grapevines. The image obtained at 07:30 was not useful because it was not possible to separate soil from canopy temperatures. Using the airborne data, the correlation between CWSI and Ψ_L had an R ² value of 0.46 at 09:30 h and of 0.71 at 12:30 h, suggesting that the latter was the more favorable time for obtaining thermal images that were linked with Ψ_L values. A sensitivity analysis of varying pixel size showed that a 0.3 m pixel was needed for precise CWSI mapping. The CWSI maps thus obtained by airborne thermal imagery were effective in assessing the spatial variability of water stress across the vineyard.     

县域尺度上基于GF-1PMS影像的冬小麦种植面积遥感监测

为探究县域尺度上基于高分一号卫星(GF-1)PMS影像进行冬小麦遥感监测的可行性及精准性,以河南省滑县为研究区,遴选2015年2月上旬GF-1 PMS影像6景,对影像进行辐射定标、FLAASH大气校正、NNDiffuse融合、几何精校正、地图投影转换等预处理后,在外业调查和样本分析的基础上构建一种新的冬小麦决策树分类模型,模型第1层决策方案中NDVI0.311的像元为冬小麦,得到冬小麦的粗分类结果;在此基础上进行第2层决策分类,以进一步提高冬小麦的分类精度,分类方案为第1波段地表反射率0.146、第2波段地表反射率0.148、第3波段地表反射率0.135、第4波段地表反射率0.250的像元为冬小麦。对分类结果进行形态学滤波处理,以消除或减少分类结果中孤立的像元。分别基于决策树分类模型与ENVI软件自带的IsoData非监督分类模型,对比分析GF-1PMS影像和同时期Landsat-8OLI影像在冬小麦面积提取上的精度。结果表明:基于新构建的决策树分类模型,2015年滑县冬小麦种植面积为115 715.81hm2,混淆矩阵检验总体精度为99.62%,Kappa系数为0.99;PMS影像提取冬小麦的混淆矩阵总体精度比OLI影像高出9个百分点。说明县域尺度上基于单时相GF-1PMS影像在冬小麦收获前提取冬小麦种植面积是可行的,提取精度较高。     

结合灰度纹理和支持向量机分类的三江源草地信息提取

以三江源地区地形地貌特征、草场分布较为典型的班玛县为例,以HJ环境星多光谱影像为主要数据,基于支持向量机SVM超平面理论,结合灰度共生矩阵寻找最适宜的分类核函数,选取了三江源地区草地信息提取的最适宜SVM分类模型,并与传统的监督分类方法最大似然法和SVM分类方法进行比较,进行三江源草地分类方法的优化。结果表明,与传统监督分类方法相比,除Sigmoid核函数外,其余结合方法的分类精度均有所提高,其中结合纹理和高斯核函数的SVM分类模型有着较理想的识别效果,精度达到91%,Kappa系数为0.856 0,能为三江源地区草地可持续利用以及生态系统恢复提供基础数据。     

Analysis of hyperspectral scattering images using locally linear embedding algorithm for apple mealiness classification

Hyperspectral scattering images between 600 nm and 1000 nm were acquired for 580 ‘Delicious’ apples for mealiness classification. A locally linear embedding (LLE) algorithm was developed to extract features directly from the hyperspectral scattering image data. Partial least squares discriminant analysis (PLSDA) and support vector machine (SVM) were applied to develop classification models based on the LLE, mean-LLE and mean spectra algorithms. The model based on the LLE algorithm achieved an overall classification accuracy of 80.4%, compared with 76.2% by the mean-LLE algorithm and 73.0% by the mean spectra method for two-class classification (i.e., mealy and nonmealy) coupled with PLSDA. For the SVM models, the LLE algorithm had an overall classification accuracy of 82.5%, compared with 79.4% by the mean-LLE algorithm and 78.3% by the mean spectra method. Hence, the LLE algorithm provided an effective means to extract hyperspectral scattering features for mealiness classification.