Search

Yield mapping for citrus fruit is a challenging task due to factors such as varying illumination conditions, clustering, and occlusions. Mapping the yield for immature citrus fruit presents an additional challenge that is the colours of fruit and leaves are almost identical. Commonly used machine vision techniques using colour cameras become less effective for immature citrus fruit detection. This study explores a novel active thermal imaging method to tackle the problem of colour similarity between immature citrus fruit and leaves. In this study, a thermal camera was combined with a water spray system that applied water mist to citrus trees. The water mist caused temperatures of both the fruit and leaf surfaces to change but at different rates. Multiple parameters of the spray system were experimented with the goal to induce as much temperature differences as possible between fruit and leaf surfaces. The combined system was tested in a citrus grove for fruit detections. Deep learning models were built based on the active thermal imaging system and tracking and fruit counting algorithms were created to count fruit in thermal videos. A mean average precision of 87.2% was achieved by the models and an accuracy of 96% was achieved when comparing the number of fruit counted by the algorithms with the true number of fruit counted manually in the field.

Fruit detection and localization will be essential for future agronomic management of fruit crops, with applications in yield prediction, yield mapping and automated harvesting. RGB-D cameras are promising sensors for fruit detection given that they provide geometrical information with color data. Some of these sensors work on the principle of time-of-flight (ToF) and, besides color and depth, provide the backscatter signal intensity. However, this radiometric capability has not been exploited for fruit detection applications. This work presents the KFuji RGB-DS database, composed of 967 multi-modal images containing a total of 12,839 Fuji apples. Compilation of the database allowed a study of the usefulness of fusing RGB-D and radiometric information obtained with Kinect v2 for fruit detection. To do so, the signal intensity was range corrected to overcome signal attenuation, obtaining an image that was proportional to the reflectance of the scene. A registration between RGB, depth and intensity images was then carried out. The Faster R-CNN model was adapted for use with five-channel input images: color (RGB), depth (D) and range-corrected intensity signal (S). Results show an improvement of 4.46% in F1-score when adding depth and range-corrected intensity channels, obtaining an F1-score of 0.898 and an AP of 94.8% when all channels are used. From our experimental results, it can be concluded that the radiometric capabilities of ToF sensors give valuable information for fruit detection. This work was partly funded by the Secretaria d’Universitats i Recerca del Departament d’Empresa i Coneixement de la Generalitat de Catalunya, the Spanish Ministry of Economy and Competitiveness and the European Regional Development Fund (ERDF) under Grants 2017SGR 646, AGL2013-48297-C2-2-R and MALEGRA, TEC2016-75976-R. The Spanish Ministry of Education is thanked for Mr. J. Gené’s predoctoral fellowships (FPU15/03355). We would also like to thank Nufri and Vicens Maquinària Agrícola S.A. for their support during data acquisition, and Adria Carbó for his assistance in Faster R-CNN implementation.

Accurate measurements of maize yields at field or subfield scales are useful for guiding agronomic practices and investments and policies for improving food security. Data on smallholder maize systems are currently sparse, but satellite remote sensing offers promise for accelerating learning about these systems. Here we document the use of Google Earth Engine (GEE) to build “wall-to-wall” 10 m resolution maps of (i) cropland presence, (ii) maize presence, and (iii) maize yields for the main 2017 maize season in Kenya and Tanzania. Mapping these outcomes at this scale is extremely challenging because of very heterogeneous landscapes, lack of cloud-free satellite imagery, and the low quantity of quality ground-based data in these regions. First, we computed seasonal median composites of Sentinel-1 radar backscatter and Sentinel-2 optical reflectance measures for each pixel in the region, and used them to build both crop/non-crop and maize/non-maize Random Forest (RF) classifiers. Several thousand crop/non-crop labels were collected through an in-house GEE labeler, and thousands of crop type labels from the 2015–2017 growing seasons were obtained from various sources. Results show that the crop/non-crop classifier successfully identified cropland with over 85% out-of-sample accuracy in both countries, with Sentinel-1 being particularly useful for prediction. Among the cropped pixels, the maize/non-maize classier had an accuracy of 79% in Tanzania and 63% in Kenya. To map maize yields, we build on past work using a scalable crop yield mapper (SCYM) that utilizes simulations from a crop model to train a regression that predicts yields from observations. Here we advance past approaches by (i) grouping simulations by Global Agro-Environmental Stratification (GAES) zones across the two countries, in order to account for landscape heterogeneity, (ii) utilizing gridded datasets on soil and sowing and harvest dates to setup model simulations in a scalable way; and (iii) utilizing all available satellite observations during the growing season in a parsimonious way by using harmonic regression fits implemented in GEE. SCYM estimates were able to capture about 50% of the variation in the yields at the district level in Western Kenya as measured by objective ground-based crop cuts. Finally, we illustrated the utility of our yield maps with two case studies. First, we document the magnitude and interannual variability of spatial heterogeneity of yields in each district, and how it varies for different parts of the region. Second, we combine our estimates with recently released soil databases in the region to investigate the most important soil constraints in the region. Soil factors explain a high fraction (72%) of variation in predicted yields, with the predominant factor being soil nitrogen levels. Overall, this study illustrates the power of combining Sentinel-1 and Sentinel-2 imagery, the GEE platform, and advanced classification and yield mapping algorithms to advance understanding of smallholder agricultural systems.

In the United States average adoption rates have increased for precision agriculture (PA) technologies used to produce many field crops. PA makes use of information collected on the farm to target site-specific, intensive management of farm production. The United States Department of Agriculture (USDA) Agricultural Resource Management Survey (ARMS) allows close examination of regional patterns of adoption, and how crop types and region interact with differences in farm sizes and soil productivity variability to influence adoption rates. The most common PA technologies are guidance systems that use global positioning systems (GPS) to steer tractors and other farm equipment. Remote sensing, soil mapping, and yield mapping all use GPS to geolocate data and create maps used to guide farm management decision. Variable rate input-application technologies (VRT) make use of remote images, soil tests, yields maps and other sources of information to apply different, more precise levels of inputs in farmer’s fields. GPS guided VRT fertilization was introduced in the early 1990s and increased slowly over the last three decades. The ARMS data for winter wheat (2017), corn (2016) and soybeans (2012) showed use of VRT seeding and pesticide applications growing rapidly. The data indicated that PA technology was being used on farms across all sizes and all regions, with adoption occurring more rapidly on larger farms. VRT use on soybean farms was highest in areas of higher soil variability.

Strawberries in California have a $2 billion direct economic impact on the state; however, they are currently produced based only on uniform field management techniques. Creating yield maps of strawberries could allow for variable-rate and site-specific applications of inputs, which could improve productivity and reduce environmental pollution. This paper presents the development and application of an instrumented strawberry-picking cart for yield mapping. During manual harvest of strawberries planted on raised beds, pickers walk inside the furrows, pick fruit from the beds on both sides, and deposit them into a tray placed on a picking cart. A ‘smart’ picking cart, similar to the standard carts, has been designed and instrumented with several types of sensors including load cells, a real-time kinematic global positioning system (RTK GPS), a microcontroller, and an inertial measurement unit (IMU). This instrumented cart serves two purposes: to work in sync with tray-transporting robots during robot-aided strawberry harvesting, and to create yield maps of strawberry fields. A yield map for an approximately 300 m 2 plot of a strawberry field in Salinas, California, was generated after the cart was calibrated. During the yield-monitoring experiment, 13.5 trays, each with a capacity of about 4.2 k g of strawberries, were filled with fruit. The mean prediction accuracy of the mass of full trays measured by the load cells was calculated to be 4.8 % .

A machine vision algorithm was developed to detect passion fruits and identify maturity of the detected fruits using natural outdoor RGB-D images. As different passion fruits on the same branch can be in different maturity stages, detection and maturity classification on a complex background are very important for yield mapping and development of intelligent mobile fruit-picking robots. In this study, a Kinect sensor was used for data acquisition, and maturity stages of the fruits were divided into five categories: young (Y), near-young (NY), near-mature (NM), mature (M) and after-mature (AM). The algorithm involved two stages. First, by colour and depth images, passion fruits were detected using faster region-based convolutional neural networks (Faster R-CNN), and colour-based detection was integrated with depth-based detection for improving detection performance. Second, for each detected fruit region, the dense scale invariant features transform (DSIFT) algorithm combined with locality-constrained linear coding (LLC) was used to extract and represent the features of fruit maturity from R, G, and B channels, respectively. In addition, the RGB-DSIFT-LLC features were input into a linear support vector machine (SVM) classifier for identifying the maturity of fruits. By conducting an experimental study on a special dataset, we verified that the proposed method achieves 92.71% detection accuracy and 91.52% maturity classification accuracy.

Citrus fruit detection is one of the most important and challenging steps in citrus yield mapping. The distinct color differences between the ripe fruit and leaves allowed previously-described imaging-based methods to achieve good results. However, immature green citrus fruit detection, which aims to provide valuable information for citrus yield mapping at earlier stages is much more difficult because the fruit and leaf colors are very similar. This study combines color and thermal images for immature green fruit detections. Experiments identified optimal conditions for thermal imaging. A multimodal imaging platform was built to integrate color and thermal cameras. A novel image registration method was developed for combining color and thermal images and matching fruit in both images which achieved pixel-level accuracy. A new Color-Thermal Combined Probability (CTCP) algorithm was created to effectively fuse information from the color and thermal images to classify potential image regions into fruit and non-fruit classes. Algorithms were also developed to integrate image registration, information fusion and fruit classification and detection into a single step for real-time processing. An increase in recall rate from 78.1% when using only color images to 90.4% after fusing the color and thermal images was obtained at similar precision rates, and an increase in precision rate from 86.6% to 95.5% was obtained at similar recall rates. The fusion of the color and thermal images effectively improved immature green citrus fruit detection.

Damage to agricultural crops due to water-logging is becoming more frequent in this era of global climate change. In this study, we investigated the effect of water-logging on winter wheat growth (LAI and aboveground biomass) and yield at a parcel scale using high spatial resolution satellite data. For this purpose, a dynamic mapping of winter wheat LAI and biomass during growth under water-logging conditions was implemented. In addition, yield mapping and yield loss estimation were also carried out to arrive at the potential impacts of water-logging on both winter wheat growth and grain yield. Two varieties of winter wheat were investigated (Yang Mai 14 and Yang Mai 18) at tillering, jointing to booting and flowering. Field conditions were representative of water-logging treatment (contrast check (CK), waterlogging (WL) and flooding (FL)). A total of eight high spatial resolution optical satellite images, including WorldView-2, WorldView-3, Pleiades-1A, SPOT-6, GeoEye-1 and SPOT-7 were obtained at time points concurrent with eight temporal field observations. Seven widely used vegetation indices (VIs) were utilized as independent variables for LAI, biomass and yield estimation. Optimal estimation models were selected according to coefficient of determination (R2), correlation coefficient (R) and root mean square error (RMSE) for subsequent mapping of LAI, biomass and yield. Results show that the power model, having EVI as the independent variable, had the highest accuracy for biomass estimation (R2 of 0.78 and RMSE of 0.076 kg/m2) while the exponential model with NDVI is the optimal model for LAI estimation (with 0.74 in R2 and 0.64 in RMSE). GNDVI is the optimal VI for yield mapping, with 0.563 in Rcv (cross validated R) and 843.76 kg/hm2 in RMSEcv (cross validated RMSE). As for the assessment of water-logging damage on winter wheat growth, FL showed a more serious impact than that of WL, based on results of the dynamic mapping of biomass and LAI at the same development stages. Water-logging from jointing to booting showed a more serious damage impact with respect to biomass and yield than that at the tillering and flowering stages. Our results have demonstrated the potential of multi-source and high spatial resolution remotely-sensed optical data for quantifying the impact of water-logging on winter wheat growth and grain yield at parcel scales, a research direction that has rarely been reported in agricultural remote sensing literature. This study thus provides a fundamental basis and the possibility for crop yield loss evaluation, water-logging monitoring, and for the design of mitigation strategies that would benefit even small-holder and subsistence farmers.

Hemp (Cannabis sativa L.) is a multipurpose industrial crop which is mainly cultivated in temperate regions. With its high potential for economic returns for its seeds and fiber, there is growing interest in cultivating hemp in many territories including Malaysia and other Asian countries, where its cultivation is currently illegal. To date, no comprehensive study on the suitability of this crop under Malaysian conditions has been conducted. In this paper, we propose an assessment framework as a roadmap to develop the hemp industry in Malaysia and possibly other Asian countries with equatorial climates. This framework includes suitability assessment (climate and soil), crop modelling (current and future yields under climate change) and economic analysis (net present value (NPV), NPV benefit (NPVB) and benefit-cost ratio (BCR). The land suitability assessment classified hemp as an adaptable crop for most of the land in the country. The AquaCrop model, parameterised from secondary data collected from literature was used in simulations and potential yield mapping. The estimated average potential seed and fiber yield at six locations between 2010 and 2019 was 1.61 ± 0.25 and 2.78 ± 0.39 t ha–1 respectively. Using five general circulation model (GCM) simulations, yields under future climates in Malaysia showed an increase in most of the locations. The highest NPVB of 1641 USD ha–1 (BCR of 1.33) for seed was estimated under current climate conditions. Yields of 1.38 t ha–1 (seed) and 3.62 t ha–1 (fiber) are the minimum economically feasible yields with a Benefit-Cost Ratio of 1.00 suggesting a potential for hemp cultivation in comparison to countries with established hemp industries. The present framework could be used to develop a pathway for adoption of hemp as a crop for the future in tropical countries.

Apart from socioeconomic factors, success of robotics in agriculture lies in developing economically attractive solutions with efficiency comparable to that of the humans. Fruit localization is one of the building blocks in many robotic agricultural operations (e.g., yield mapping and robotic harvesting) that determines 3D Euclidean positions of the fruits using one or several sensors. It is crucial to guarantee the performance of the localization methods in the presence of fruit detection errors and unknown fruit motion (e.g., due to wind gust), so that the desired efficiency of the subsequent systems can be achieved. For instance, inaccurate localization may severely affect fruit picking efficiency in robotic harvesting. The presented estimation-based localization approach provides estimates of the fruit positions in the presence of fruit detection errors and unknown fruit motion, and it is based on a new sensing procedure that uses multiple (⩾2) inexpensive monocular cameras. A nonlinear estimator called particle filter is developed to estimate the unknown position of the fruits using image measurements obtained from multiple cameras. The particle filter is partitioned into clusters to independently localize individual fruits, while the behavior of the clusters is manipulated at global level to maintain a single filter structure. Since the accuracy of localization is affected by errors in fruit detection, the presented sensor model includes non-Gaussian fruit detection errors along with image noise. Fruit motion can significantly reduce harvesting efficiency due to errors in locating moving fruits. In contrast to existing methods, the dynamics of fruit motion are derived and included in the localization framework to obtain time-varying position estimates of the moving fruits. A detailed theoretical foundation is provided for the new estimation-based fruit localization approach, and it is validated through extensive Monte Carlo simulations. The performance of the estimator is evaluated by varying the design parameters, measurement noise, number of fruits, amount of overlap in clustered fruit scenarios, and fruit velocity. Correlation of these parameters with the performance of the estimator is derived, and guidelines are presented for selecting the design parameters and predicting performance bounds under given operating conditions.

Yield mapping techniques have only recently started to be implemented by the Californian wine grape industry, but the advancement has necessitated new processing methods for large vineyards. The process for mapping large blocks harvested with multiple machines has only recently occurred and implies that their yield monitors have to be calibrated and corrected to the same scale. Here we discuss two methods for processing yield maps at the commercial level. Method 1 depends on many calibrations with delivered fruit weight to a winery. Method 2 normalizes raw files automatically can reduce total processing time by up to 90%.

High-resolution yield maps are an essential tool in modern agriculture. Using spatial interpolation, spatially discrete sampled yield data from yield monitors can be transformed into continuous yield maps. However, spatial interpolation is usually performed using methods that can be computationally demanding or that lack credibility measurements. The objectives of this work were to improve and evaluate a spatial machine learning algorithm for yield mapping at a fine scale. The core method used for mapping is Quantile Regression Forest Spatial Interpolation (QRFI), in which covariates from the spatial neighborhood of the sampled yields are used to predict yields at unsampled sites. To assess the algorithm performance, more than one thousand yield monitor datasets from several plant species were processed with QRFI, and other geostatistical (ordinary kriging, KG) and non-geostatistical (spatial inverse distance interpolation, IDW) methods. We illustrated the application of QRFI for yield mapping using yield monitor datasets of different grain crops from the Argentine Pampas. Evaluation of the methods showed that all statistical metrics suggested better results for yield maps obtained by QRFI than by KG or IDW. Globally, prediction error of QRFI was 11.5%, which was on average at least 16% better than the corresponding results of the other spatial interpolation methods. The machine learning algorithm QRFI can be successfully applied to perform spatial interpolation of yields at the field scale and to assess the associated prediction uncertainty.

Crop yield maps estimated from satellite data increasingly are used to understand drivers of yield trends and variability, yet satellite-derived regional maps are rarely compared with location-specific yields due to the difficulty of acquiring sub-field ground truth data at scale. In commercial agricultural systems, combine harvesters with onboard yield monitors collect real-time yield data during harvest with high spatial resolution, generating a large ground dataset. Here, we leveraged a yield monitor dataset of over one million maize field observations across the United States Corn Belt from 2008 to 2018 to evaluate the Scalable Crop Yield Mapper (SCYM). SCYM uses region-specific crop model simulations and climate data to interpret vegetation indices from satellite observations, thus enabling efficient sub-field yield estimation across large regions and multiple years without reliance on ground data calibration. We used the ground dataset to compare alternative SCYM model implementations, define minimum required satellite observation criteria, and evaluate the sensitivity of satellite-based yield estimates to on-the-ground variation in management, soil, and annual weather. We found that smoothing annual time series data with harmonic regression increased 30 m pixel-level accuracy from r2 = 0.31 to 0.40 and reduced dependency on specific satellite observation timing, enabling robust yield estimation on 97% of annual maize area using only Landsat data. Agreement improved as the assessment was aggregated to field-level (r2 = 0.45) and county-level (r2 = 0.69) scales, demonstrating the need for fine-resolution ground truth data to better assess sub-field level accuracy in high resolution products. We found that SCYM and ground data showed a similar yield response to management and environmental variation, particularly capturing linear and nonlinear responses to sowing density, soil water holding capacity, and growing season precipitation. However, sensitivity to factors like soil quality and planting date was muted for SCYM estimates compared to ground-based yields. Random forest models were able to match SCYM performance when trained on at least 1000 ground observations, but performed poorly when tested on years and locations not represented in the training data. Our results demonstrate that satellite yield maps can provide much-needed information on multidecadal yield trends and inform yield gap analyses.

A cloud-based Harvest Management Information System (HMIS) has been developed.It credits pickers for the fruit they have harvested without impeding the process.The HMIS provides real-time tracking of harvest, yield mapping, and traceability.It simplifies data analysis process and visualizes the harvest and labor data.It improves accuracy of payroll and enhances decision making and logistics. The harvest process for specialty crops is generally one of intensive activity because many people are required for harvest and packing, and the harvest window is brief due to the high perishability of the produce. Herein we present a cloud-based Harvest Management Information System (HMIS) that combines a novel real-time Portable Labor Monitoring System (PLMS) with a cloud-based harvest management software. The PLMS comprised of three key elements (1) a self-leveling scale, (2) electronic control box, and (3) a frame that supports all hardware. The electronic control box includes: (i) a RFID reader, (ii) a LCD display, (iii) a thermal printer, (iv) a GPS module, and (v) a communication system. RFID tags, containing unique ID numbers, embedded within rubber wrist bands, are worn by pickers. This system can read a picker's ID (RFID bracelet), measure the weight of fruit, and record the time and location (optional) of every fruit 'transaction' (i.e., every time a picker brings a bucket of fruit to the collection bin). The collected data can be transmitted wirelessly to the server in real-time. The cloud-based software receives and processes the PLMS data on labor activities, visualizes the collected data, and can extract the data necessary for management information and automated filling of documents (e.g. payroll, yield maps). The HMIS is unique in its ability to: (1) accurately credit pickers for the fruit they have harvested in the field without impeding or altering the harvest process, (2) streamline data entry to payroll, (3) provide real-time tracking of harvest, yield mapping, and traceability, and, (4) generate precise and reliable harvest efficiency data. This integrated system was evaluated in sweet cherry, blueberry and apple orchards in Washington, USA. The weight of harvested fruit, time and location of every fruit drop were calculated accurately; all the data were transmitted wirelessly to the server and no errors were recorded.

Real-time and accurate on-branch fruit recognition in an uncontrolled/unstructured environment of orchards could facilitate Precision Horticulture (PH) practices. These practices which are based on the site-specific or variety-specific treatment of an orchard include applications like remote recognition of tree species, variety-specific orchard agrochemical applications, orchard yield mapping, robotic fruit picking, fruit tree disease treatment, etc. For this purpose, in the current work a Convolutional Neural Network (CNN) was developed and optimized for fruit recognition based on RGB images. The images from six classes of on-branch fruits i.e. green apples, nectarine, apricot, peach, sour cherry, and amber-colored plums were captured from local orchards at Semnan province, Iran. To avoid over-fitting, the dataset was then augmented to create more training data from existing samples. The model consisted of multiple convolutional, max-pooling, Global Average Pooling (GAP), and fully connected layers. Using GAP instead of the Flatten layer improved the performance of the model by increasing the accuracy as well as significantly reducing the trainable parameters (about 65 times reduction). The optimization phase of the model development was performed by testing four optimizers (RMSprop, SGD, Adam, and Nadam) with three batch sizes (16, 32, and 64) each with 50 epochs. Accordingly, Nadam optimizer with batch size = 32 demonstrated the best results. The best configuration achieved the accuracy of 99.8% and the cross-entropy loss of 0.019 for the test dataset. This result shows that the model is well developed and has good generalization. This reflects the potential of the method for the remote recognition and classification of different varieties of fruits in an orchard regardless of the environmental effects like complex background, variable light, overlaps, and occlusions with other plant parts, etc. The proposed network was also compared with popular structures like VGG11, ResNet50, ResNet152, and YOLOv3. The processing time of this model was about 8 ms per image while it was 351 ms for ResNet152, proving that the proposed network is much better for real-time applications. Consequently, this study presents a robust method for fulfilling the requirements of a PH practice in a high-tech horticulture system.

Lack of yield mapping solutions is currently a bottleneck for Precision Agriculture development and adoption in many manually harvested fruit and vegetable crops. In such systems, the handpicked produce is briefly stored in bags or boxes across the field before they are loaded and transported. This study tested a simple yield mapping method based on georeferencing the bags used during harvest with local yield calculated based on the distribution of these points across the field. Virtual yield maps and real field data were used to validate different data processing methods under different scenarios; scenarios included different levels of yield spatial variability and bag positioning and mass errors. Method 1 calculated yield at each bag point by estimating the area needed to fill it; such area was based on the bag distance to its neighbours. Method 2 calculated local yield based on the distribution of bags across an area using a moving window approach. In normal field situations – with bag positioning and mass errors below 1 m and 5% – the approaches had similar performance with accuracy levels varying between 5 and 11 Mg ha−1, depending on the yield spatial variability. With increasing bag positioning error, method 2 outperformed method 1. Both approaches were little affected by error in bag mass estimation. Overall, the yield mapping methods are useful in supporting most applications in Precision Agriculture and can be easily implemented in a software tool to promote user adoption and site-specific management.

Fruit detection and localization are essential for future agronomic management of fruit crops such as yield prediction, yield mapping and automated harvesting. However, to perform robust and efficient fruit detection and localization in orchard is a challenging task under variable illumination, low-resolutions and heavy occlusion by neighboring fruits, foliage, or branches. Therefore, researches of fruit detection and localization by getting more information of objects are essential. RGB-D (Red, Green, Blue -Depth) cameras are promising sensors and widely used in fruit detection and localization given that they provide depth information and infrared information in addition to RGB information. After presenting a discussion on the advantages and disadvantages of RGB-D cameras with different depth measurement principles and application fields, this paper reviews various types of RGB-D sensor systems and image processing methods used for fruit detection and localization in the field. Finally, major challenges for the successful application of RGB-D camera-based machine vision system, and potential future directions for the research and development in this area are discussed.

Forecasts of crop yield are an important tool for a variety of stakeholders, but most studies produce large-scale, late-season yield forecasts that are not appropriate for farmers. Farmers require forecasts of crop yield mid-season and at fine spatial resolutions to guide site-specific adaptive crop management practices. This study created empirical random forest models to forecast cotton yield mid-season at three different spatial resolutions: 30 m, field, and farm aggregation. Large yield mapping datasets from 14 different seasons were utilised to build the model across 68 different fields from six large cotton farms in eastern Australia. A generic conceptual framework is proposed to empirically model crop yield. This topsaw framework represents the spatial and temporal factors that drive yield, including topography (t), organisms (o), plant measurements (p), soil (s), agronomy/management (a), and weather (w). In this study, the specific predictor variables to represent these driving factors of yield included elevation (t), Normalised Difference Vegetation Index, Enhanced Vegetation Index, MODIS Evapotranspiration (p), gamma radiometrics grid, clay content digital soil map (s), rainfall data, and growing day degrees (w). Forecasts of cotton yield increased in accuracy as the spatial resolution of predictions became coarser. Using a leave-one-year-out cross-validation, cotton yield could be forecasted to an accuracy of 0.42 Lin's Concordance Correlation Coefficient (LCCC) and Root Mean Square Error (RMSE) of 2.11 bales ha−1 at 30 m resolution. This improved to a 0.63 LCCC and 1.72 bales ha−1 RMSE at the field resolution, and a 0.65 LCCC and 0.77 bales ha−1 at the aggregation-scale. As the developed approach solely relied on publicly available predictor variables, there is an opportunity to apply this to any cotton field in Australia, and over a much larger area. Overall, this study is an important step in building an operational approach to forecast mid-season cotton yield at fine spatial resolutions, and has the potential to improve production, input use efficiency, and profitability for individual growers and the Australian Cotton Industry as a whole.

An accurate and real-time image based multi-class fruit detection system is important for facilitating higher level smart farm tasks such as yield mapping and robotic harvesting. Robotic harvesting can reduce the costs of labour and increase fruit quality. This paper proposes a deep learning framework for multi-class fruits detection based on improved Faster R-CNN. The proposed framework includes fruits image library creation, data argumentation, improved Faster RCNN model generation, and performance evaluation. This work is a pioneer to create a multi-labeled and knowledge-based outdoor orchard image library using 4000 images in the real world. Also, improvement of the convolutional and pooling layers is achieved to have a more accurate and faster detection. The test results show the proposed algorithm has achieved higher detecting accuracy and lower processing time than the traditional detectors, which has excellent potential to build an autonomous and real-time harvesting or yield mapping/estimation system.

The use of high spatial resolution, on-the-go proximal soil sensing of apparent electrical conductivity (ECa) through electromagnetic induction (EMI) is increasingly common, in concert with yield mapping, to assist in the delineation of management zones for Precision Agriculture (PA). Less common, but gaining in popularity, is the use of gamma-radiometric (γ) soil sensing. Using contrasting sites in South Australia and Queensland, the specific objectives of the study were to assess for each site, region or all sites together, how well soil cation exchange capacity (CEC) and clay content may be predicted by EMI and γ sensing; to see whether the predictions were improved when both sensors were used, compared to a single sensor; and to evaluate the potential utility of the multi-sensor data in terms of understanding the variation in observed crop yield within sites. Of particular interest was evaluating a generic, as opposed to site-specific, approach to the simultaneous use and calibration of EMI and γ sensing at contrasting sites chosen across a dispersed geography and pedology. EMI and γ soil surveys were carried out at five sites across three cereal growing regions in South Australia, and at three sites in Queensland used for sugarcane production. Soil samples were also collected from each site for laboratory analysis. Data analysis comprised simple correlation analysis between soil sensor data and soil properties; fusion of sensor data by region and across all sites using weighted principal component analysis (PCA), with the data weighted on the basis of the two source sensors (weight of 0.5 assigned to ECa and the remaining weight divided equally amongst 238U, 232Th, 40K and ‘total count’ (CPS); weights of 0.125 to each). The output from the PCA was used to predict maps of CEC and clay using multiple regression. Simple correlation analysis showed the expected potential utility of both sensors for predicting soil properties by site and by region. The first three principal components (PCs) explained 98% of the data variation across regions and all sites. Models for the prediction of CEC and clay content, derived from the all sites PCs, were significant (p

Many sensing systems were developed for use in precision agriculture in Florida.Examples include fruit recognition and disease detection platforms.A summary of past efforts is presented.Applications of different sensing systems and future directions are discussed. Many sensing systems have been developed for use in precision agriculture in Florida over the past decade. These systems have been designed for specialty crops such as citrus and blueberry. Systems include those for fruit recognition for yield mapping as well as those for disease detection systems using ground- and aerial-based platforms. Other systems discussed are used in soil phosphorus detection using near-infrared (NIR) and Raman spectroscopy, debris detection generated from citrus mechanical harvesting, detection of citrus fruit dropped on the ground due to disease, citrus leaf nitrogen detection, silage yield mapping, soil nutrients and grain insect detection using NIR spectroscopy. A summary of past efforts is presented in this paper, applications of these different sensing systems are discussed, and future directions are described.

Outdoor hyperspectral images of blueberry bushes were taken and analyzed.Hyperspectral image was shown to be able to separate fruit stages and background.PWCD, HDR and NG measure were used for band selection.Selected bands performed well in classifying fruit stages and background. Hyperspectral imagery divides spectrum into many bands with very narrow bandwidth. It is more capable to detect or classify objects, where visible information is not sufficient for the task. However, hyperspectral image contains a large amount of redundant information, which eliminates its discriminability. Band selection is used to both reduce the dimensionality of hyperspectral images and save useful bands for further application. This study explores the feasibility of hyperspectral imaging for the task of classifying blueberry fruit growth stages and background. Three information theory based band selection methods using Kullback-Leibler divergence: pair-wise class discriminability, hierarchical dimensionality reduction and non-Gaussianity measures were applied. Three classifiers, K-nearest neighbor, support vector machine and AdaBoost were used to test the performance of the selected bands by the three methods. The selected bands achieved classification accuracies of 88% and higher. Therefore, the band selection methods are very useful in reducing the volume of the hyperspectral data, and constructing a multispectral imaging system for detecting blueberry fruit maturity stages.

We identify the key yield-limiting factors of soil properties and stages of NDVI measured respectively for tobacco.We characterize the spatial variability of soil properties and vegetation index.We delineate management zones using soil properties and NDVI measurements respectively by use of fuzzy clustering algorithm.It is feasible to use an active canopy sensor to delineate MZs instead of using soil properties. Recent precision agriculture research has focused on the utilization of management zones (MZs) as a method for variable-rate fertilizer applications. This study tries to make use of an active canopy sensor to define MZs in contrast with the ones using soil properties. To achieve this objective, a 20-m grid-sampling scheme was imposed on the field with 101 points, and we collected soil samples from the top 20cm at each point before tobacco planting in April 2013. The normalized difference vegetative index (NDVI) data at five growth stages of the tobacco growth cycle were measured by using a GreenSeeker handheld crop sensor at the location of each sample point. Yield mapping was carried out at harvest. Through stepwise multiple analysis, we identified the key yield-limiting factors of soil properties and stages of NDVI measured respectively. For soil property data, OM, AP and Fe were used for yield-based MZs while for canopy parameters, NDVI during knee-high and flowering were utilized. Fuzzy c-means clustering algorithm was performed in order to determine the optimal number of the clusters. Results showed that the optimal number of MZs with regard to the two methods were both five and variance analysis indicated the heterogeneity of soil properties and yield among them. Consequently, it is feasible to use an active canopy sensor to delineate management zones for tobacco-planting fields.

This study was conducted to identify blueberry fruit of different growth stages using natural outdoor images toward the development of a blueberry yield mapping system. As blueberries usually contain different maturity stages in a same branch, identification of blueberry fruit and their maturity stages from different background is very important for yield mapping. In this study, maturity stages of the fruit were divided into four categories: mature (m), near-mature (nm), near-young (ny) and young (y). A stepwised algorithm, termed 'color component analysis based detection (CCAD)' method, was developed and validated to identify blueberry fruit using outdoor color images. Firstly, a dataset was built using manually cropped pixels from training images. Three color components, red (R), blue (B) and hue (H), were selected using the forward feature selection algorithm (FFSA), and used to separate all fruit of four maturity stages from background through different classifiers. In this work, not only the traditional classifiers such as K-nearest neighbor (KNN), and naive Bayesian classification (NBC) were used, but another newly introduced 'supervised K-means clustering classifier (SK-means)' was also developed and applied to the dataset. In the second step, classifiers were built to separate a group of 'mature & near-mature' fruit from a group of 'near-young & young' fruit from all fruit pixels. Finally, classifiers were developed to separate mature fruit from near-mature fruit, and near-young fruit from young fruit. The classifiers obtained from these different steps were then applied to validation images, resulting in final identification. Cross validation was conducted using these different classifiers and their results were compared. KNN classifier yielded the highest classification accuracy (85-98%) from the validation set of the prebuilt pixel dataset collected from the training images in all separations. An one-way ANOVA was used to compare the performance of the three classifies, which shows KNN performed significantly better than other methods. The newly proposed 'SK-means' classifier yielded a fairly high accuracy (90%) for the separation of mature and near-mature fruit. The newly developed 'CCAD' method for blueberry was proved to be efficient for identifying blueberry fruit of different growth stages using natural outdoor color images toward the development of a blueberry yield mapping system.

Yield mapping for tree crops by mechanical harvesting requires automatic detection and counting of fruits in tree canopy. However, partial occlusion, shape irregularity, varying illumination, multiple sizes and similarity with the background make fruit identification a very difficult task to achieve. Therefore, immature green citrus-fruit detection within a green canopy is a challenging task due to all the above-mentioned problems. A novel algorithmic technique was used to detect immature green citrus fruit in tree canopy under natural outdoor conditions. Shape analysis and texture classification were two integral parts of the algorithm. Shape analysis was conducted to detect as many fruits as possible. Texture classification by a support vector machine (SVM), Canny edge detection combined with a graph-based connected component algorithm and Hough line detection, were used to remove false positives. Next, keypoints were detected using a scale invariant feature transform (SIFT) algorithm and to further remove false positives. A majority voting scheme was implemented to make the algorithm more robust. The algorithm was able to accurately detect and count 80.4% of citrus fruit in a validation set of images acquired from a citrus grove under natural outdoor conditions. The algorithm could be further improved to provide growers early yield estimation so that growers can manage grove more efficiently on a site-specific basis to increase yield and profit.

Quantification of the gap between water-limited yield and actual crop yields at a sub-field scale across grain farms are now becoming possible in industrialised agriculture because of routine yield mapping and readily available methods to estimate potential yield. Analyses at sub-field scale can be used to target intervention strategies such as matching fertiliser inputs to yield variation or amelioration of soil constraints. As such they complement the predominant “top-down” approaches to the quantification of yield gaps and are relevant to individual farm decision making. In this study we bring together yield maps, farm soil surveys, and rainfall records to generate whole-farm maps of actual yield, water-limited potential yield, the yield gap, the contribution of soil constraints to the yield gap, and a list of the feasible interventions to close that gap. We apply the approach in a case study of a 4500 ha rainfed farm, located in the low-rainfall cropping zone of the Western Australian grainbelt. We found a substantial gap between water-limited potential yield and actual yield, in an environment where wheat yields typically range between 0.5 and 2.5 t/ha. On a whole-of-farm basis, mean wheat yield was 50–60% of the water-limited yield and the yield gap varied from 0.6 t/ha in a dry year to 1.5 t/ha in wet year. The size of the yield gap varied among and within soil types due to associated soil constraints, and the analysis suggested that addressing the predominant constraint of acidity on the average performing areas on the farm could generate the greatest gains, rather than focussing on the below-average areas that have large yield gaps, which are afflicted by intractable soil constraints. The approach can be used by advisors and farmers to assess the proportion of the farm that reaches attainable or water-limited yield, and a tool to target regions within the farm that fail to reach attainable yield. This style of analysis lends itself to landscapes where there is a strong association between crop yield and soil type and environments where yield is strongly determined by rainfall and hence water-limited yield can be easily calculated. With an increase in the routine use of yield monitors by grain growers we anticipate an increase in yield map coverage across farms that will eventually make our method readily applicable.

Highlights? High spatial variability of soil, yield and quality in apples. ? High temporal variability over the 3years of the experiment. ? Management zones created using multivariate geostatistics. Management zone (MZ) delineation is important for the application of Precision Agriculture because farm management decisions are based on it. Several factors were used for the MZ delineation including crop and soil characteristics. In the present paper multivariate analysis was applied to delineate MZs. Soil and crop data, collected over 3years from a Precision Agriculture project in an apple orchard in Greece, were used. The collected data were categorized in three groups, namely soil properties, yield and fruit quality. All data were analyzed for descriptive statistics and their distribution. Maps of the spatial variability for the 3years were presented. Data were jointly analyzed for management zone delineation using a combination of multivariate geostatistics with a non-parametric clustering approach, and the orchard was divided in four zones which could be differently managed. However, further research and experimentation are needed before precision horticulture being confidently adopted in Greece.

A machine vision system for estimation of citrus fruit mass, fruit count, and fruit size during postharvest processing was investigated with the aim of developing of an advanced citrus yield mapping system. Such yield mapping system enables the citrus growers to efficiently manage the in-grove spatial variability factors such as: soil type, soil fertility, moisture content, etc., and can help increase yield and profits. Thus, a machine vision system was developed and installed in a citrus debris cleaning machine, which removes debris from mechanically harvested loads. An image processing algorithm was developed to identify fruit from images of the postharvest citrus from a commercial citrus grove. For fruit detection, logistic regression model based pixel classification algorithms were developed. To avoid misclassification due to highly saturated area on fruit and non-fruit regions, a highly saturated area recovering (HSAR) algorithm was developed that obviated use of a “filling holes” operation. A mass calibration process was conducted, and fruit mass was estimated, which turned out to be reasonably good. The highest coefficient of determination (R2) value between the measured fruit mass and the estimated fruit mass was observed to be 0.945 and the root mean square error was 116.1 kg. A H-minima transform based watershed algorithm was used to separate the joined fruit and enabled an estimation of fruit counting and fruit size. Fruit mass estimation using the fruit size information was also conducted and these results were compared with that of the mass estimation based on fruit pixel area.

Wild blueberry fruit yield maps could be used to generate prescription maps for site-specific application of fertilizer to improve crop productivity. An automated yield monitoring system II (AYMS II) consisting of two @mEye color cameras, real time kinematics-global positioning system, custom software, and a ruggedized laptop computer was developed and mounted on a Specialized Farm Motorized Vehicle (SFMV) for real-time fruit yield mapping. Custom software was developed in C++ programming language to acquire and process image in real-time from both cameras and store the percentage of blue pixels in ruggedized laptop computer. Two wild blueberry fields were selected in central Nova Scotia to evaluate the performance of the AYMS II for fruit yield mapping. Calibration was carried out at 20 randomly selected data points in each field. The ripe fruit was hand-harvested out of a 2.0x0.7m quadrat at each selected point and camera images were also taken from the same points to calculate the percentage of blue pixels (fraction of blue pixels in the image). Linear regression analysis was performed to calibrate the actual fruit yield with the percentage of blue pixels. Real-time yield mapping was carried out with AYMS II. The estimated yield along with geo-referenced coordinates was imported into ArcGIS 9.3 software for detailed mapping. The AYMS II, hardware and software performed well to estimate wild blueberry fruit yield. The linear regression results showed highly significant relationship between the percentage of blue pixels and actual fruit yield in field 1 and field 2. The correlation between actual and predicted fruit yield (validation, using the equation from field 2) in field 1 and field 2 (validation, using the equation from field 1) was also highly significant. Maps developed in ArcGIS 9.3 showed substantial variability in fruit yield in both fields. The bare spots were coincided with no or low yielding areas in the fields. The viable AYMS II will be incorporated into wild blueberry harvester. This information obtained from AYMS II could be used to implement site-specific management practices within the blueberry fields to optimize productivity while minimizing the environmental impact of farming operations.

The non-destructive assessment of forage mass in legume-grass mixtures as a tool for yield mapping in precision farming applications has been investigated in two field experiments. An ultrasonic sensor was used to determine sward heights. Forage mass-height relationships were evaluated by carrying out static measurements on binary legume-grass mixtures of white clover (Trifolium repens L.), red clover (Trifolium pratense L.), and lucerne (Medicago sativa L.) with perennial rye grass (Lolium perenne L.) across a wide range of sward heights (5.0-104.2cm) and forage mass (0.15-11.25tha^-^1). Mobile measurements, hereafter referred to as ''on-the-go'' were conducted by mounting the ultrasonic sensor in combination with a high-precision Differential Global Positioning System (DGPS) on a vehicle. Data were recorded along experimental plots consisting of perennial rye grass and grass-clover mixtures similar to the mixtures that were used for the static experiment. The static experiment revealed a relationship between ultrasonic sward height and forage mass explaining 74.8% of the variance with a standard error (SE) of 1.05tha^-^1 in a common dataset. The type of legume species, weed proportion, and growth period had a significant impact on the above mentioned relationship. Legume-specific regression functions had higher R^2-values of up to 0.855 (white clover mixture). Datasets including legume-specific mixtures and pure swards of both components reached comparable R^2 values between 0.799 and 0.818 but exhibited higher SE values. The abundance of weeds resulted in increased ultrasonic sward heights for the same levels of forage mass. On-the-go measurements across experimental field plots yielded a sward height range of 1.4-70.4cm. Abrupt forage mass changes at the transition from treatment plots to cut interspaces resulted in a significant deviation from stubble height within a distance of 50cm to plot borders. When legume-specific equations derived from static measurements were applied to sward heights, forage mass was overestimated by 21.4% on average. Mean residuals from predicted forage mass ranged between 0.893 (pure grass) and 1.672 (red clover mixture) and increased significantly if the point sampling distance along the track was increased to more than 0.82m on average across all plots. The prediction accuracy of forage mass from ultrasonic height measurements is promising; however, further modifications to the technique are necessary. One such improvement can be the use of spectral reflectance signatures in combination with the ultrasonic sensor.

The ability to map profit across a cotton field would enable producers to determine where money is being made or lost on their farms and to implement precise field management practices to facilitate the highest return possible on each portion of a field. Mapping profit requires knowledge of site-specific costs and revenues, including yield and price. Price varies site-specifically because fiber quality varies, so mapping fiber quality is an important component of profit mapping. To map fiber quality, the harvest location of individual cotton bales must be known, and thus a system to track the harvest location of cotton modules must be available. To this end, a wireless module-tracking system was recently developed, but automation of the system is required before it will find practical use on the farm. In Part 1 of this report, research to develop automatic triggering of wireless messages is described. In Part 2, research to enable the system to function with multiple harvesting machines of the same type in the same field - a common situation in commercial cotton farming - is described along with testing of the entire automated wireless module-tracking system. To automate wireless-message triggering, a sensing and control system was added to a harvester to indicate when the machine is dumping a basket load of cotton so that wireless messages can be automatically sent from the harvester to subsequent field machines. This automated system was incorporated into the existing wireless module-tracking system, field tested, and it ultimately operated as designed, without human intervention. Linking data collected with this system together with cotton classing data enabled the creation of fiber-quality maps.

The ability to map profit across a cotton field would enable producers to determine where money is being made or lost on their farms and to implement precise field management practices to facilitate the highest return possible on each portion of a field. Mapping profit requires knowledge of site-specific costs and revenues, including yield and price. Price varies site-specifically because fiber quality varies, so mapping fiber quality is an important component of profit mapping. To map fiber quality, the harvest location of individual cotton bales must be known, and thus a system to track the harvest location of cotton modules must be available. To this end, a wireless module-tracking system was recently developed, but automation of the system is required before it will find practical use on the farm. In Part 1 of this report, research to develop automatic triggering of wireless messages is described. In Part 2, research to enable the system to function with multiple harvesting machines of the same type in the same field - a common situation in commercial cotton farming - is described along with testing of the entire automated wireless module-tracking system (WMTS). An RFID system was incorporated, and it enabled the WMTS to correctly and consistently differentiate among various harvesting vehicles. The improved WMTS subsequently sent wireless messages to the correct machines when cotton transfers were made in the presence of multiple harvest machines. Overall testing proved that the automated WMTS worked largely as designed. When both complete and partial cotton basket dumps were simulated, the correct wireless-messaging decision was made 100% of the time.

Wild blueberry (Vaccinium angustifolium Ait.) is a unique crop native to Northeastern North America with annual production of 82,000 t from an area of 79,000 ha. Yield maps along with topography and soil nutrient maps could be used to develop precise site-specific nutrition programs for wild blueberry production. Here, the development of an automated yield monitoring and mapping system for real-time fruit yield estimation for wild blueberry is presented. The hardware components of the system consist of a digital colour camera, a positioning system receiver and a computer mounted on a specially designed farm motorised vehicle. The yield prediction method was based on the estimation of the blue pixels representing ripe fruit in the field of view of each image and their expression as a percentage of total image pixels, using it as the correlation parameter with the corresponding yield. The image processing software was development in “Delphi” and “C” programming languages. Field experiments were carried out in two commercial wild blueberry fields. Concerning the calibration of the system, the experimental results show a significant correlation between percentage blue pixels and actual fruit yield (R2 = 0.90; P

Calibrating cotton yield monitor data is difficult, because there are many confounding factors, such as landscape, soil, ambient light and temperature, defoliation level, and variety that can cause sensor response to shift from field to field and even within a field. The conventional calibration method involves in situ weighing of basket loads, which is costly, logistically unfavorable, and not adequate to account for inconsistent performance of yield monitor sensors. In this work a module-specific post-processing method was used to calibrate cotton yield data. The method was enabled by a wireless communication and GPS based electronic system that can delineate the harvest location of each cotton module in the field during machine harvest. Module-specific correction coefficients were calculated as the ratio of module weight determined at the gin to the yield monitor based module weight integrated over the module harvest area. Module-specific lint turnout, also determined at the gin, was used to convert seed cotton yield to lint yield. The method was field tested on 16 cotton modules harvested from two study areas near College Station, TX. Results showed that yield monitor based weight deviated substantially from the actual weight for some modules, with errors ranging from -41.0% to +13.9%. Lint yield maps produced with module-specific post-processing calibration showed more reliable spatial patterns than maps produced with field-based post-processing. Combining the wireless communication and GPS system with the cotton yield monitor could greatly enhance the quality of cotton yield maps, making them more suitable for various precision agriculture applications.

In this paper two methods are proposed for automatically matching bins containing harvested fruits with corresponding trees, during harvesting in orchards, where GPS data may be unavailable due to foliage. Both methods use a long-range radio frequency identification (RFID) antenna located on the harvesting platform for tree identification. Bin registration is accomplished in the first method by passive RFID tags attached to the bins, whereas the second method uses a barcode reader located on the platform, and low-cost barcode tags on the bins. Additionally, a digital scale is used with both methods to measure the yield distribution in the field, during the loading of the bins. An experimental evaluation of these methods was performed during peach and kiwi harvesting in two different fields in Northern Greece. The aim was to estimate the tree and bin detection accuracies of both methods and their effect on the bin loading time. Statistical analysis of the data showed that when compared to the current standard harvesting procedure, RFID bin registration did not affect the amount of time to stack a bin on the platform (loading time), whereas barcode bin registration increased this time by 14%. It was also found that the use of the particular scale increased the loading time by almost 33% in both bin registration methods. Finally, the detection accuracy for the trees was 100% in all experiments and for the bins it was almost 100% for the RFID and 100% for the barcode reader. The results suggest that barcode technology can be used reliably for bin registration, without delaying the harvesting. Tree detection with long-range RFID technology was reliable; however tree growth combined with other factors such as wind, sunlight, etc., might decrease the tree detection accuracy over long periods of time. Finally, the bins had better be weighed at the packinghouse in order to generate the yield map, unless a much faster scale can be used in the field.

Recent advances in on-the-go soil sensing, terrain modelling and yield mapping have made available large quantities of information about the within-field variability of soil and crop properties. But the selection of the key variables for an identification of management zones, required for precision agriculture, is not straightforward. To investigate a procedure for this selection, an 8 ha agricultural field in the Loess belt of Belgium was considered for this study. The available information consisted of: (i) top- and subsoil samples taken at 110 locations, on which soil properties: textural fractions, organic carbon (OC), CaCO3 and pH were analysed, (ii) soil apparent electrical conductivity (ECa) obtained through an electromagnetic induction based sensor, and (iii) wetness index, stream power index and steepest slope angle derived from a detailed digital elevation model (DEM). A principal component analysis, involving 12 soil and topographic properties and two ECa variables, identified three components explaining 67.4% of the total variability. These three components were best represented by pH, ECa that strongly associated with texture and OC. However, OC was closely related to some more readily obtainable topographic properties, and therefore elevation was preferred. A fuzzy k-means classification of these three variables produced four potential management classes. Three-year average standardized yield maps of grain and straw showed productivity differences across these classes, but mainly linked to their landscape position. In the loess area with complex soil-landscape interactions pH, ECa and elevation can be considered as key properties to delineate potential management classes.

In the light of recent advances in spectral imaging technology, highly flexible modeling methods must be developed to estimate various soil and crop parameters for precision farming from airborne hyperspectral imagery. The potential of artificial neural networks (ANNs) for the development of in-season yield mapping and forecasting systems was examined. Hyperspectral images of corn (Zea mays L.) plots in eastern Canada, subjected to different fertilization rates and various weed management protocols, were acquired by a compact airborne spectral imager. Statistical and ANN approaches along with various vegetation indices were used to develop yield prediction models. Principal component analysis was used to reduce the number of input variables. Greater prediction accuracy (about 20% validation RMSE) was obtained with an ANN model than with either of the three conventional empirical models based on normalized difference vegetation index, simple ratio, or photochemical reflectance index. No clear difference was observed between ANNs and stepwise multiple linear regression models. Although the high potential usefulness of ANNs was confirmed, particularly in the creation of yield maps, further investigations are needed before their application at the field scale can be generalized.

Yield mapping is a major component of precision farming. In contrast to the yield mapping available on the market in combine harvesters, yield mapping in forage harvesters is still in the research and development stage. The principles known from the specialist literature are based either on measuring the mass flow or the volume flow. The subject of this paper is improvement of the measuring accuracy of the principle of measuring the width of the gap between the feed rolls of a forage harvester by taking into account the material behaviour of selected typical forage crops (forage rye, pasture grass, spring barley, silo maize). On the basis of compaction experiments under defined laboratory conditions, regression equations are drawn up which can be applied to describe the material behaviour for the pressure area from 0 up to 3·5 bar. The result is a universally applicable algorithm for more accurate throughput measurement, which can be realized in the on-board computers of modern forage harvesters.

One of the latest trends on commercial combine harvesters is online mapping of grain yield. Together with sensors registering the surface harvested each instance and a position system, a grain flow measurement is installed at the end of the threshing process. Due to this position, the filtering impact of the machine smoothes the mass flow variations, weakening the economic value of the constructed maps. Here, a method is proposed for reducing this drawback based upon an earlier constructed grain flow model. The results are presented from a wheat field fertilized with different doses of nitrogen and sowed with different crop densities.

A simple system has been developed to generate yield maps of hand-harvested citrus in a reliable, low-cost manner using a commercial GPS recorder. Individual 0.7-m3 container locations are mapped to indicate yield variations within a citrus block. Further future refinements are discussed. The system is applicable to other manually harvested crops and harvesting systems.

Yield mapping, to date, has been achieved principally by the use of grain flow measuring systems in conjunction with ground positioning systems (GPS) that can be permanently fitted to combine harvesters. Mobile or tractor mounted GPS systems are now becoming available so presenting the opportunity to use different sources of information for crop yield data. A yield mapping system has been developed for use in non-grain crops (i.e. roots and forage) using the measurement of mass accumulation rate. The concept was realised by instrumenting a high-sided trailer to allow the rate of accumulation of crop mass to be recorded as the trailer follows the harvester. The location of the trailer is recorded using a differential GPS with a linked data set to log the position and the processed mass information. To remove unwanted noise due to trailer pitch, roll and yaw, and the tractor engine vibrations, the crop mass data are conditioned using a combination of analogue signal filtering and subsequent numerical processing. The results of field harvests for which yield maps were produced give confidence that the principle of mass accumulation rate can be used in a range of harvesting machines and trailers. A yield map obtained for a forage crop showed a mean yield of 32.8 t/ha with a range from 26 to 35 t/ha. The mean yield values from the yield map and from samples measured by batch harvesting were not significantly different. Yield maps for single crops of sugar beet and potatoes showed yields between 11 and 68 t/ha (mean 49 t/ha) and between 10 and 50 t/ha (mean 35.9 t/ha), respectively. For the sugar beet crop there was no significant difference between the mean yield values for batch weighed samples and recorded yield map values.

The application of precision agriculture to cotton (Gossypium hirsutum L.) production has been limited due to the lack of a suitable yield monitor. Cotton yield monitors under research have utilized load cells, sound sensors and electro-optical devices to measure seedcotton yield. A commercial electro-optical cotton yield sensor was introduced to the cotton industry in 1997 by the Zycom Corporation, Bedford, MA. The purpose of this study was to evaluate the precision of this new yield monitor on small research plots. Cotton research plots were harvested and monitor values recorded from a single harvesting head. Seedcotton was saved from each plot and weighed for comparison with yield monitor values. Monitor values were compared to cumulative plot weights. A total of 226.8 kg of seedcotton was harvested from 72 plots. A significant linear relationship (r2=0.99) was observed between monitor values and observed plot weights. This yield monitor looked very promising for estimating cotton yields and suggests that this technology may allow for the application of precision agriculture to cotton production.

A field-level geographic information system (FIS) is being developed. The system provides special analytical functions useful for research in precision farming. Spatial data from various sources with different accuracies and resolutions are integrated. Remotely sensed data are used to assist field-data analysis. Region blocking and partitioning are obtained using logic and arithmetic queries and overlays. Spatial patterns are analyzed in spatial and frequency domains in both Cartesian and polar coordinate systems. A discrete model was developed to simulate the time lag inherent to a yield sensor on a combine harvester. Z-transformation was used to numerically remove the lagging effect.

Most yield mapping systems on combine harvesters measure the rate of clean grain flow. However, this rate at any instant is a function of yield over a finite region along the swath. This smoothing effect will limit the spatial resolution of the yield map. As a simplification it is assumed that flow rate is a convolution of yield along the swath and an impulse response function (i.r.f.) which summarises all processes in the combine. A model of this i.r.f. was derived from flow data for the opening and closing sections of 35 passes. The results were found to be consistent with an independent estimate of the time delay for grain flow through the combine and field comparisons of yield variation and flow rate for adjacent swaths. For the combine used, spatial resolution would seem to be limited to 15 to 25 m. Deconvolution to achieve finer resolution appeared to be unsuccessful, probably because of noise in the flow sensor.

Spatially variable field operations depend on various parameters such as in-field variations of soil and crop yield. A yield map can provide local information on nutrient absorption, variability of soil and effects of special treatment strategies. In order to monitor the spatial variations of yield the combine harvester has to be equipped at least with data acquisition for locating and measuring grain flow and area productivity. Accuracy of yield mapping can be increased by additional measuring of grain moisture, actual cutting width and the dynamics of the grain transport within the combine. This paper comments on special technical problems encountered during yield mapping on a combine and presents the technical equipment. The suitability of the sensors has been examined in laboratory and field tests.

This contribution reports investigations on yield detection of grain with combines. It describes the two types of yield measurement systems used, the position detection with the Global Positioning System (GPS) and the data acquisition system. Information on the precision attained is given by analyses. The data evaluation leads to yield mapping for individual plots and for the entire experimental operation. Raster and contour graphics were prepared.

Yield mapping may form an important part of an in-field site-specific crop production system, both for the spatial analysis of production efficiency and for the determination of spatially optimized application rates for fertilizer and sowing of seed. The objective of the work reported here, was to apply the principle of yield mapping to whole crop harvesting corn silage. The yield measurement is based on the continuous recording of the material flowrate through harvesting machine, the machine driving speed and the machine location in the field. For the flowrate measurement, the shaft of the material blower and the drive shaft of the base unit powering the cutterhead, feedrolls and front attachement, were instrumented with strain gauge torque transducers. Within the calibration range, the signals of these sensors showed a linear relationship to the flowrate. The harvester was also equipped with a speed radar and a data acquisition system based on a personal computer. The location tracking of the machine was done by integrating the machine speed and manual recording of the machine path in the field. The construction of a yield map, from the recorded signals, included several digital signal processing operations. The resulting map of a 1·2ha corn silage hield showed spatial yield variation from 1·2 to 4·8 kg/m 2 with average of 3·2 kg/m 2 and standard deviation of 0·64 kg/m 2 .