Multilevel systematic sampling to estimate total fruit number for yield forecasts

Early forecasting of fruit orchard yield is important for market planning and for growers and exporters to plan labour, bins, storage and purchase of packing materials. Large variations in tree yield pose a challenge for accurate yield estimation. We evaluated a three-level systematic sampling procedure for unbiased estimation of fruit number for yield forecasts. In the Spring of 2009 we estimated the total number of fruit in several rows of each of 14 commercial fruit orchards growing apple (11 groves), kiwifruit (two groves), and table grapes (one grove) in central Chile. Survey times were 10–100 min for apples (depending on vigour), 85 min for the table grapes, and 85 and 150 min for the kiwifruit. During harvest in the Fall, the fruit were counted to obtain the true number. Yields ranged from lows of several thousand (grape bunches), to highs of more than 40 000 fruit (apples, kiwifruit). Absolute true errors (defined as the absolute difference between the estimate and the true value, divided by the true value) were less than 5% in six orchards, between 5 and 10% in a further five orchards and 13% in one orchard. In two apple orchards we obtained absolute true errors of about 20%. Error analysis based on systematic sub-sampling across each sampling stage was used to determine how to distribute sampling effort to achieve a total coefficient of error of 10%. We discuss the extension of the procedure for yield estimation at the full orchard scale for any target precision.     

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.     

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.     

Spatial variation in yield and quality in a small apple orchard

We describe the yield and quality of apples from a 0.8 ha apple orchard located in northern Greece over two growing seasons and consider the potential for site-specific management. The orchard has two apple cultivars: Red Chief (main cultivar) and Fuji (pollinator). Yield was measured by weighing all fruit harvested from groups of five adjacent trees and the position of the central tree was recorded by GPS. Apple quality at harvest was evaluated from samples of the two cultivars in both years for which fruit mass, flesh firmness, soluble solids content, juice pH and acidity of the juice were determined. The variation in tree flowering was also measured in the spring of the second season using a stereological sampling procedure. The results showed considerable variability in the number of tree flowers, yield and quality across the orchard for both cultivars. The number of flowers was strongly correlated with the final yield. These data could potentially be used to plan precise thinning and for early prediction of yield; the latter is important for marketing the fruit. Several quality characteristics, including fruit juice soluble solids content and acid content were negatively correlated with yield. The general patterns of spatial variation in several variables suggested that changes in topography and aspect had important effects on apple yield and quality.     

基于遥感的高山松连清固定样地地上生物量估测模型构建

  目的  研究利用遥感方法构建高山松固定样地地上生物量估测的参数模型,可以在今后前期样地的基础上直接快速、准确地估测生物量,或者开展少量的外业调查即可获取地上生物量。  方法  基于遥感因子与样地地上生物量变化量和线性混合模型提高生物量估测精度,以香格里拉市1987、1992、1997、2002、2007、2012、2017年7期国家森林资源清查固定样地和对应年份Landsat TM、OLI的Level-1数据为基础,首先对遥感数据进行预处理:包括辐射定标、大气校正、几何校正和地形校正,提取原始波段、比值因子、植被指数、图像增强信息、纹理指数、混合像元分解后的丰度、叶面积指数,计算5 ~ 30年间隔样地对应的遥感因子变化值。根据森林资源二类调查的高山松分布特征,选择地形因子作为线性混合模型的固定和随机效应,采用多元线性回归、非线性回归、地理加权回归、线性混合模型构建高山松地上生物量估测的静态模型,基于遥感光谱信息变化量构建了有树高和无树高参与的动态模型。最后对不同的建模方法和验证结果进行对比分析,选择最优结果作为估测模型并验证。  结果  （1）分析静态数据建模和验证的结果,采用样地号为固定因子、坡度等级为随机因子的线性混合模型的拟合R2最高,为0.75;但利用训练数据集和2017年数据验证,其精度都较低。（2）分析变化量数据建模和验证的结果,采用样地号为固定因子、坡度等级为随机因子、遥感因子变化量为自变量的线性混合模型拟合R2最高,为0.70,预测精度P值为（68.86 ± 11.93）%;增加平均树高变化量,拟合R2最高为0.79,预测P值为（73.39 ± 6.18）%。（3）无论是有、还是无树高参与的变化量模型其拟合和预测精度都达到80%,其预测精度达到了非参数模型预测精度。  结论  基于变化量的估测模型的拟合和预测精度较静态模型有所提高;综合遥感因子、地形因子构建的高山松地上生物量估测线性混合模型,其精度有较大提高;采用遥感因子变化量构建的高山松地上生物量估测模型,有效弥补了静态光学遥感数据估测生物量的不足,经检验可用于其他年期的估测。      

Airborne Hyperspectral Imagery and Yield Monitor Data for Mapping Cotton Yield Variability

Increased availability of hyperspectral imagery necessitates the evaluation of its potential for precision agriculture applications. This study examined airborne hyperspectral imagery for mapping cotton (Gossypium hirsutum L.) yield variability as compared with yield monitor data. Hyperspectral images were acquired using an airborne imaging system from two cotton fields during the 2001 growing season, and yield data were collected from the fields using a cotton yield monitor. The raw hyperspectral images contained 128 bands between 457 and 922 nm. The raw images were geometrically corrected, georeferenced and resampled to 1 m resolution, and then converted to reflectance. Aggregation functions were then applied to each of the 128 bands to reduce the cell resolution to 4 m (close to the cotton picker's cutting width) and 8 m. The yield data were also aggregated to the two grids. Correlation analysis showed that cotton yield was significantly related to the image data for all the bands except for a few bands in the transitional range from the red to the near-infrared region. Stepwise regression performed on the yield and hyperspectral data identified significant bands and band combinations for estimating yield variability for the two fields. Narrow band normalized difference vegetation indices derived from the significant bands provided better yield estimation than most of the individual bands. The stepwise regression models based on the significant narrow bands explained 61% and 69% of the variability in yield for the two fields, respectively. To demonstrate if narrow bands may be better for yield estimation than broad bands, the hyperspectral bands were aggregated into Landsat-7 ETM+ sensor's bandwidths. The stepwise regression models based on the four broad bands explained only 42% and 58% of the yield variability for the two fields, respectively. These results indicate that hyperspectral imagery may be a useful data source for mapping crop yield variability.     

用PLS算法由HJ-1A/1B遥感影像估测区域冬小麦理论产量

【目的】作物遥感估产是遥感技术在农业生产中研究与应用的重点领域,能够向大田区域生产提供及时可靠的产量信息,准确地估测作物产量,对于确保国家粮食安全,制定社会发展规划,指导和调控宏观种植业结构调整,提高涉农企业与农民的经营管理水平具有重要意义,为进一步提高遥感估产精度,显示国产影像在农业估产中的应用效果。通过筛选冬小麦理论产量的敏感遥感变量,构建基于国产影像的理论产量遥感估测模型,实现区域冬小麦理论产量遥感估测,为及时了解不同生态区域冬小麦产量丰欠变化趋势提供参考。【方法】以2010年4月26日、2011年4月28日、2012年4月28日和2013年5月2日冬小麦开花期四景HJ-1A/1B影像为遥感数据,提取出13个遥感变量,以江苏省泰兴、姜堰、仪征、兴化、大丰5县作为试验采样区,于各实验区选取具有代表性的样点进行采样,并于室内进行测定,将335个实测的冬小麦理论产量样本按3﹕2比例分成建模集和验证集样本,依据估算残差平方和处于最小值确定模型所需主成分数,将决定系数、均方根误差和相对误差为模型评价参数,利用建模集样本分析了卫星遥感变量与冬小麦理论产量的定量关系,运用偏最小二乘回归算法构建及验证了以理论单产为目标的多变量遥感估产模型,将其算法模型估产效果与线性回归算法和主成分分析算法模型进行比较,并制作了冬小麦理论产量空间等级分布图。【结果】理论产量与所选的大多数遥感变量间关系密切,且多数遥感变量两两间具有极显著的多重相关性;理论产量偏最小二乘回归模型的最佳主成分数为4,且结构加强色素植被指数、归一化植被指数、绿色归一化植被指数和植被衰减指数为理论产量遥感估测的敏感变量;经建模集和验证集评价,理论产量估测模型的决定系数分别为0.79和0.76,均方根误差分别为720.45和928.05 kg·hm^-2,相对误差分别为11.45%和13.92%,且估测精度比线性回归算法分别提高了25%以上和27%以上,比主成分分析算法分别提高了15%以上和16%以上,说明偏最小二乘回归算法模型估测区域理论产量的效果明显好于线性回归和主成分分析算法,且具有较强的应用能力。【结论】该模型应用结果与冬小麦理论产量实际区域分布情况相符合,为提高遥感对区域冬小麦理论产量的估测精度提供了一种有效途径,有利于大面积应用和推广。     

Software to quantify and map vegetative cover in fallow fields for weed management decisions

Mapping weed cover during the fallow period of dryland crop rotations would be valuable for weed management in subsequent crops and could be done with low cost color digital cameras, however most managers lack the specialized software and expertise needed to create a map from the images. A system of software was developed to quantify weed cover in fallow fields in digital images and to simplify and automate the most challenging tasks that non-GIS professionals confront in creating and using maps derived from a large number of images. A GIS file of image locations is created with inexpensive consumer software. Images are classified, a GIS file is generated and the map is displayed in a simple GIS viewer with free software we developed. A map can be generated from 1000 images and 5000 GPS coordinates in 30 min, including image classification. The classified and original images for all locations can be viewed together easily from the map application. The accuracy of estimating weed cover was evaluated using images collected in 15 fields under natural light with a consumer grade camera mounted on an ATV driving 8-11 km h⁻¹. Weed cover was estimated with 96% accuracy for images, regardless of the amount of crop residue, unless part of the image was shaded by the camera. In those images, accuracy was 90% or better. This system will work with many professional and consumer digital cameras and GPS units and the classification algorithm can be easily modified for other applications.     

基于无人机RGB影像估测田间小麦穗数

单位面积穗数是小麦产量构成的重要因素,利用图像信息处理技术快速、准确地估测田间小麦穗数,可以为小麦长势监测和产量估测提供直接依据.利用无人机路径规划和控制系统(fragmentation monitoring and analysis with aerial photography,FragMAP)获取标准统一、高分辨...     

基于HJ小卫星影像的北京市冬小麦测产研究

粮食产量的估测对制定粮食政策和调整种植结构具有重要的意义。本研究以北京市为例,利用2013年4月30日、5月12日和5月29日三期冬小麦抽穗、灌浆时期的HJ小卫星NDVI数据,结合北京市实产地块数据建立了北京市冬小麦估产回归模型,并对北京市各区县和北京全市的冬小麦单产进行了估算。结果显示:分区县地块回归的北京市冬小麦主产区大兴、房山、通州和顺义的小麦单产分别为5148.96 kg/hm2、4849.30 kg/hm2、5350.64kg/hm2和5108.84 kg/hm2,P值分别为0.000、0.000、0.000和0.001;北京市全市的冬小麦单产为5049.24 kg/hm2,精度验证的结果显示实测单产和预测单产具有良好的对应关系,其R2=0.92,RE=2.18%,RMSE=154.61 kg/hm2。以上研究结果表明利用HJ小卫星的NDVI数据可以快速、准确的估算北京市及其各区县的冬小麦单产。     

苹果分层纺锤形冠层结构与果实产量品质的关系

【目的】研究分层纺锤形冠层结构与产量品质关系,分析树形结构的适宜性,为短枝苹果整形修剪提供依据。【方法】以分层纺锤形苹果树为研究对象,测定冠层微气候因子、枝量、冠层结构及产量和品质参数,分析比较不同冠层间的差异。【结果】不同冠层的光照、温度、ACF、DIFN、MTA都随冠层高度的增加而增加;湿度,主枝长,主枝粗,冠层体积、LAI、产量随冠层高度的增加而减小;而果实品质仅有果实着色面积在上层与中、下层上存在显著差异,而其它品质指标在各层间无显著差异,果实品质一致性较高;产量主要集中于树体中下部冠层,便于修剪、树体管理、采摘。【结论】树冠下部冠层产量平均为下部冠层产量平均为40 392 kg/hm²,中部冠层产量为15 081 kg/hm²,上部冠层产量为8 316 kg/hm²,产量主要集中于冠层中下部,便于管理和采摘,苹果分层纺锤形冠层通风透光条件良好,产量较高,品质指标在各层无显著差异,果实品质一致,有较好生产应用价值。     

基于GF-1PMS影像的森林郁闭度定量估测模型研究

森林郁闭度是我国森林资源二类调查中的重要林分因子之一,为探索高分一号影像在森林郁闭度定量估测的可行性,以内蒙古自治区东北部柴河林业局为研究区,基于GF-1 PMS多光谱影像和DEM数据,以纹理特征、光谱信息和地形因子为自变量,采用k-最近邻法（k-Nearest Neighbor,k-NN）、稳健估计以及偏最小二乘法3种方法构建研究区域的森林郁闭度估测模型,辅之以二类调查小班数据和现地实测数据进行模型精度评价。结果表明:1）3种郁闭度估测方法的应用效果均能满足实际需求,这说明GF-1 PMS多光谱影像在森林郁闭度定量估测方面具有一定的潜力;2）k-NN法和稳健估计的实际应用效果明显优于偏最小二乘法,且2种模型的估测精度均>80%;3）3种郁闭度估测稳定性分析,结果显示k-NN法稳定性较好,而稳健估计法和偏最小二乘法估测模型不稳定。     

扁桃-冬小麦间作模式下树体结构、光环境和小麦产量间的相关性

【目的】研究果粮间作模式下,扁桃树体结构指标与间作区域光环境指标、光环境指标与间作冬小麦产量构成指标的相关关系,为新疆南疆果粮间作模式下冬小麦产量的预估、高光效树形的选择和优化提供理论依据。【方法】以扁桃-冬小麦(新冬20号)间作模式为研究对象,测定不同树形树体结构指标、间作区域光环境指标和小麦产量构成指标;分析产量指标与光环境指标的相关性、树体结构指标与间作区域光环境指标的相关性、树体结构指标与扁桃负载量的相关性。【结果】新疆南疆地区扁桃-冬小麦间作模式下,小麦单位面积有效穗数、穗粒数和千粒重分别与拔节期、扬花期和灌浆期PAR强度呈极显著正相关关系;小麦千粒重和单穗粒重与灌浆期日平均光照强度、及400~1 400 μmol/(m²·s)PAR持续时长呈极显著正相关关系;树体负载量与树冠体积、树冠投影面积和平均冠幅等指标呈极显著正相关关系;冠高与树体西侧冠下株间区光照指标显著相关,冠幅对树体西侧冠下株间区和远冠区2个区域光照指标显著相关。【结论】灌浆期400~1 400 μmol/(m²·s) PAR的持续时长和冠幅可以作为新疆南疆地区果粮间作模式下,小麦产量预估、评级及高光效树形筛选的主要评价指标。     

基于GIS的河南省烤烟移栽面积遥感监测及产量估算

烟叶的移栽面积和产量信息对于有效配置烟叶资源实现供求平衡非常重要,通过利用多时相的可见光数据对2012年河南省烟叶播种面积和长势进行了卫星遥感监测,同时对河南省各地市烟叶产量进行了估测。经过验证,估测面积与实测面积高度一致,估测精度达到90%。移栽面积估测误差主要来源于烟田种植面积细碎化和遥感影像分辨率较粗。面积估测误差势必会影响产量估测精度,此外由于缺少长时间数据积累不能准确估计叶面积指数和产量之间的数量关系,也会带来产量的估测误差。     

A new algorithm for automatic Rumex obtusifolius detection in digital images using colour and texture features and the influence of image resolution

In Gebhardt et al. (2006) an object-oriented image classification algorithm was introduced for detecting Rumex obtusifolius (RUMOB) and other weeds in mixed grassland swards, based on shape, colour and texture features. This paper describes a new algorithm that improves classification accuracy. The leaves of the typical grassland weeds (RUMOB, Taraxacum officinale, Plantago major) and other homogeneous regions were segmented automatically in digital colour images using local homogeneity and morphological operations. Additional texture and colour features were identified that contribute to the differentiation between grassland weeds using a stepwise discriminant analysis. Maximum-likelihood classification was performed on the variables retained after discriminant analysis. Classification accuracy was improved by up to 83% and Rumex detection rates of 93% were achieved. The effect of image resolution on classification results was investigated. The eight million pixel images were upscaled in six stages to create images with decreasing pixel resolution. Rumex detection rates of over 90% were obtained at almost all resolutions, and there was only moderate misclassification of other objects to RUMOB. Image processing time ranged from 45 s for the full resolution images to 2.5 s for the lowest resolution ones.     

Estimation of bioenergy crop yield and N status by hyperspectral canopy reflectance and partial least square regression

The objective of this study was to compare performance of partial least square regression (PLSR) and best narrowband normalize nitrogen vegetation index (NNVI) linear regression models for predicting N concentration and best narrowband normalize different vegetation index (NDVI) for end of season biomass yield in bioenergy crop production systems. Canopy hyperspectral data was collected using an ASD FieldSpec FR spectroradiometer (350–2500 nm) at monthly intervals in 2012 and 2013. The cropping systems evaluated in the study were perennial grass {mixed grass [50 % switchgrass (Panicum virgatum L.), 25 % Indian grass “Cheyenne” (Sorghastrum nutans (L.) Nash) and 25 % big bluestem “Kaw” (Andropogon gerardii Vitman)] and switchgrass “Alamo”} and high biomass sorghum “Blade 5200” (Sorghum bicolor (L.) Moench) grown under variable N applications rates to estimate biomass yield and quality. The NNVI was computed with the wavebands pair of 400 and 510 nm for the high biomass sorghum and 1500 and 2260 nm for the perennial grass that were strongly correlated to N concentration for both years. Wavebands used in computing best narrowband NDVI were highly variable, but the wavebands from the red edge region (710–740 nm) provided the best correlation. Narrowband NDVI was weakly correlated with final biomass yield of perennial grass (r² = 0.30 and RMSE = 1.6 Mg ha^?1 in 2012 and r² = 0.37 and RMSE = 4.0 Mg ha^?1, but was strongly correlated for the high biomass sorghum in 2013 (r² = 0.72 and RMSE = 4.6 Mg ha^?1). Compared to the best narrowband VI, the RMSE of the PLSR model was 19–41 % lower for estimating N concentration and 4.2–100 % lower for final biomass. These results indicates that PLSR might be best for predicting the final biomass yield using spectral sample obtained in June to July, but narrowband NNVI was more robust and useful in predicting N concentration.     

Using a robust variogram to find an adequate butterfly neighborhood size for one-step yield mapping using robust fitting paraboloid cones

The yield map is generated by fitting the yield surface shape of yield monitor data mainly using paraboloid cones on floating neighborhoods. Each yield map value is determined by the fit of such a cone on a neighborhood that looks like a huge butterfly flying along the harvest track. Wide wings of the butterfly guarantee that the map is sufficiently smoothed out across the tracks. The coefficients of regression for modeling the paraboloid cones and the scale parameter are estimated using robust weighted M-estimators where the weights decrease with the distance from one to zero; the latter is at the border of the selected neighborhood. The robust way of estimating the model parameters supersedes a procedure for detecting outliers. For a given neighborhood size, this yield mapping method is implemented by the Fortran program butterflymap.exe , which can be downloaded from the web. To obtain the appropriate size of the selected neighborhood, the variance of the yield map values should equal the variance of the true yields, which is the difference between the variance of the raw yield data and the error variance of the yield monitor. It is estimated using a robust variogram on data that have not had the trend removed. Based on investigating butterfly neighborhoods the yield map was optimized if the search radius across the harvest tracks was eight times the swath width. One reason for this wide neighborhood is that the regression used for modeling the paraboloid cones is based on weights that decrease linearly from 1 in the middle to zero at the border of the neighborhood, so only data points close to the middle have a large weight.     

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.     

Integrating multispectral reflectance and fluorescence imaging for defect detection on apples

This research investigated multispectral imaging to detect various defects on apples. An integrated approach using multispectral imaging in reflectance and fluorescence modes was used to acquire images of three varieties of apples. Eighteen images from a combination of filters ranging from the visible region through the NIR region and from three different imaging modes (reflectance, visible light induced fluorescence, and UV induced fluorescence) were acquired for each apple as a basis for pixel-level classification into normal or disorder tissue. Artificial neural network classification models were developed for two classification schemes, a two-class and a multiple-class. In the two-class scheme, pixels were categorized into normal or disordered tissue, whereas in the multiple-class scheme, pixels were categorized into normal, bitter pit, black rot, decay, soft scald, and superficial scald tissues. A 10-fold cross validation technique was used to assess the performance of the neural network models. The integrated imaging model of reflectance and fluorescence was effective on Honeycrisp variety, whereas single imaging models of reflectance or fluorescence was effective on Redcort and Red Delicious. The technique is promising for accurate recognition of different types of disorder on apple.