Search

Interpolation techniques are usually employed in the yield maps generation. In this article, a method based in spatial low-pass filtering is presented and compared with interpolation techniques. The method consists in the application of low-pass filters typically used for image processing. The yield map is constructed in a similar way to a Digital Elevation Model (DEM), except for the fact that heights represent crop quantities.

The presence of weeds, bare spots, and variation in fruit yield within wild blueberry fields emphasizes the need for yield mapping for site-specific application of agrochemicals. An automated yield monitoring system (AYMS) consisting of a digital color camera, differential global positioning system, custom software, and a ruggedized laptop computer was developed and mounted on a specially designed Farm Motorized Vehicle (FMV) for real-time fruit yield mapping. Two wild blueberry fields were selected in central Nova Scotia to evaluate the performance of the AYMS. Calibration was carried out at 38 randomly selected data points, 19 in each field. The ripe fruit was hand-harvested out of a 0.5- × 0.5-m quadrant at each selected point and camera images were also taken from the same points to calculate the blue pixel ratio (fraction of blue pixels in the image). Linear regression was used to calibrate the actual fruit yield with percentage blue pixels. Real-time yield mapping was carried out with AYMS. Custom software was developed to acquire and process the images in real-time, and store the blue pixel ratio. The estimated yield per image along with geo-referenced coordinates was imported into ArcView 3.2 GIS software for mapping.

Yield monitors have become an indispensable part of many precision agriculture systems because of their ability to quickly and efficiently measure yield variability within a field. The current technology of measuring grain yields with sensors mounted in the clean grain elevator is error prone due to mass flow variations caused by the threshing and separating operations. To reduce the effect of combine dynamic errors, a new crop mass flow sensing technology is being developed. This new sensor measures the tension on the feed conveyor drive chain, as it is believed that the tension on this chain is related to the flow of biomass through the feeder housing. The initial field tests were conducted in 160 m strips of corn that were broken into six blocks of varying grain yields. An analysis of the data indicated that chain tension sensors are sensitive to differences between the blocks of varying grain yields. Additional tests, under normal field conditions, were also conducted under varying biomass volumes controlled by selecting one of three specific stripper plate settings on the combine corn head. The chain tension sensor could detect differences between the stripper plate settings, as the greatest chain tensions were recorded at the narrowest stripper plate width. Results from both field trials were compared to yield data collected from an impact-style mass flow sensor located in the clean grain elevator. The results indicated that biomass flow sensing at the feeder housing might complement existing technologies to improve yield monitor data quality.

A machine vision-based citrus fruit counting system was developed for a continuous canopy shake and catch harvester. The system consisted of a 3CCD camera, four halogen lamps, an encoder, and a laptop computer. A total of 719 images were taken during an experiment on a test bench at the Citrus Research and Education Center (Lake Alfred, Fla.) and were used for analysis. The system was also tested on a canopy shake and catch harvester at a grove located in Fort Basinger, Florida, where a total of 773 images were acquired and 60 images were used for validation. Fruit weight was measured during image acquisition in 14 test bench experiments and in two field tests with a commercial canopy shake and catch harvester. An image processing algorithm that could identify fruit and measure its size was developed using a Bayesian classifier, morphological operations, and watershed segmentation. From the sets of color images, the number of fruit and total fruit areas were measured using the developed algorithm. Finally, the number of citrus fruits that were identified by the image processing algorithm during the test bench experiment was compared against actual fruit weight. The coefficient of determination (R2) between them was 0.962. To validate the canopy shake and catch harvester experiment, the number of fruit was counted manually from a total of 60 images. Density clustering was used to enhance the result of the Bayesian classifier. The manual count was compared with the image processing algorithm count. The R2 value was 0.891 between the actual and estimated number of fruit.

In this review, the process of estimating crop yield from yield monitor data is traced from the instant of harvest, through combine processes and sensing, to post-harvest processing, yield estimation and mapping. Unified notation is applied for easy comparison. Models for various components contributing to mapped yield estimates are detailed. These include approaches to determining basic properties such as combine position, gathering width and area swept out as a function of time. Special attention is given to comparison of the commonly-used deterministic flow model with several probabalistic alternatives that have been offered. Though there has been little success in using flow models to mitigate mixing effects, the results have been better for determining transport delay. Post-processing is discussed, including yield data screening as well as mapping. The recent integration of remote sensing with yield monitor data is described. In the proper context, this integration of remote sensing shows potential to increase the accuracy and spatial resolution of yield mapping.

It is important for orchard owners to be able to estimate the quantity of fruit on the trees at the various growth stages, because a tree that bears too many fruits will yield small fruits. Thus, if growers are interested in controlling the fruit size, knowing in advance that there are too many developing fruits will give them the opportunity to treat the tree. This study proposes a machine vision-based method of automating the yield estimation of apples on trees at different stages of their growth. Since one of the most difficult aspects of apple yield estimation is distinguishing between green varieties of apples or those that are green in the first stages of growth, and the green leaves that surround them, this investigation concentrates on estimating the yield of green varieties of apples. Hyperspectral imaging was used, because it is capable of giving a wealth of information both in the visible and the near-infrared (NIR) regions and thus offers the potential to provide useful results. A multistage algorithm was developed that uses several techniques, such as principle components analysis (PCA) and extraction and classification of homogenous objects (ECHO) for analyzing hyperspectral data, as well as machine vision techniques such as morphological operations, watershed, and blob analysis. The method developed was tested on images taken in a Golden Delicious apple orchard in the Golan Heights, Israel, in two sessions: one during the first stages of growth, and the second just before harvest. The overall correct detection rate was 88.1%, with an overall error rate of 14.1%.

A commercial combine equipped with a yield monitor and Global Positioning System (DGPS) was constructed to include a weighing system in the clean grain tank. The scale was able to output its current readings two times per second to a data-logger. The reading from the yield monitor and the scale could be matched to check the accuracy of the yield monitor summary as well as the individual (1 s) readings. Harvest data from fields of corn and oats were collected using this system. Data from both the scale and yield monitor were analyzed to verify the accuracy of the yield monitor. The scale was verified to be accurate by calibration. Results showed that the yield monitor has two data processing streams; one gives the individual weights (1 s interval weights) and the other gives the summary weights (total load weights). The results showed that these two streams are slightly different. Further analysis was conducted to understand these differences. The results showed an increase in the yield monitor accuracy with the harvest strip length. Data collected in 1997 were more accurate than data collected in 1996. It was concluded that improvement was primarily due to the updated yield monitor system (ver. 6.02 CJC).

A cooperative effort between researchers, manufacturers, and growers has been investigating precision farming applications in Florida citrus. Citrus yields, based on the location of volume-based containers, were mapped using a conventional fruit-loading truck, manual harvesters (> 99% of Florida citrus is manually harvested), and GIS/GPS components. These maps were overlaid on geo-referenced aerial photographs of the tree canopies. Two fruit weighing systems were mounted on a fruit-loading truck and integrated with the GIS/GPS components to investigate mapping weight-based yields. Results to date indicate the truck-mounted weighing systems were within 1 to 6% of certified scale weights on 20 t loads of fruit. Electronically recording harvester identity is being integrated with yield mapping to make the entire system more reliable and attractive to harvesters and growers.

A diaphragm-based grain flow sensor has been developed for use on a grain combine. The diaphragm sensor is similar to conventional impact flow sensors, but it uses a flexible fabric-reinforced rubber diaphragm to isolate the grain from the load cell that measures the impact force. The original geometry of the elevator delivery area is preserved; no plates or other devices protrude into the grain flow. Accuracy of the sensor was determined by comparison to a catch bin and a weighing conveyor. The relationship between impact force and grain flow was linear at high grain flows, but somewhat non-linear at low flow rates. Diaphragm life has exceeded 6800 Mg (250,000 bushels).

The measurement of actual harvested area per unit time is an important component in the creation of accurate crop yield maps. For row crops, such as corn, these measurements can be made manually on most conventional yield monitors. However, in drilled or broadcast crops a more accurate and automated method is required. In this study, a vector method was developed to determine actual combine harvest area at each time step of the harvest process from a global positioning system (GPS) trajectory. The algorithm was coded into a geographic information system (GIS), and modifications were made to increase computational efficiency. The method was compared with a previously reported raster method of harvest area determination on data collected during the 1997 drilled soybean harvest, using high accuracy real time kinematic GPS data. The vector method improved harvest area estimates by an average of 11% over the assumption of a constant, full swath width and provided a number of distinct advantages over the raster method.

In harvesting narrow row or broadcast crops such as wheat, the actual combine cut width can vary significantly within a field. Assuming a constant combine cut width in the whole field to calculate crop yields will give incorrect results. In this study, a bitmap method is developed to determine the effective combine cut width from Global Positioning System (GPS) positions of the combine in the field. The method consists of first initializing a bitmap that represents the pre-harvest crop conditions in the field, and then progressively updating the bitmap to represent the up-todate crop conditions during the harvest process. The effective combine cut width at each time step is derived by manipulating the changes of the bitmap during that time period. An example application of the method is shown with yield data for 1994 spring wheat. The average cut widths among combine passes varied from 1.18 m to 2.24 m, compared with a desired cut width of 1.52 m. The accuracy of the resulting yield maps was greatly improved by the bitmap method.

Location coordinate information is needed in precision agriculture to map in-field variability, and to serve as a control input for variable rate application. Differential global positioning system (DGPS) measurement techniques were compared with other independent data sources for sample point location and combine yield mapping operations. Sample point location could be determined to within 1 m (3 ft) 2dRMS using C/A code processing techniques and data from a high-performance GPS receiver. Higher accuracies could be obtained with carrier phase kinematic positioning methods, but this required more time and was a less robust technique with a greater potential for data acquisition problems. Data from a DGPS C/A code receiver was accurate enough to provide combine position information in yield mapping. However, distance data from another source, such as a ground-speed radar or shaft speed sensor, was needed to provide sufficient accuracy in the travel distance measurements used to calculate yield on an area basis.

Variable rate application technology based on spectral radiance has not previously been used for correcting in-season winter wheat nitrogen (N) deficiencies. Soil and yield mapping has been used to recommend variable amounts of applied fertilizer in crop production, however, both are restricted by the time required to obtain results and their utility is bound by the year in which they were generated. The objectives of this study were to determine the relationship between spectral radiance at specific wavelengths with wheat forage yield and forage N uptake, and to evaluate the potential use of spectral radiance measurements for correcting in-season wheat N deficiencies using sensor-based variable rate technology. Five studies were conducted, three in farmer fields where variable soil N deficiencies were present and two on Oklahoma Agricultural Experiment Station land. Spectral radiance readings for red and near-infrared (NIR) wavelengths were obtained in wheat between Feekes physiological stages 4 and 6 using photodiode-based sensors fitted with interference filters and interfaced to an embedded microcontroller. Correlation between a plant nitrogen spectral index (PNSI), a variation of the normalized-difference-vegetative-index (NDVI), and total N uptake in wheat forage was then established. Based on the PNSI readings, a variable 0 to 112 kg N ha–1 topdress N rate was determined for 3 ¥ 3 m plots and N as urea ammonium-nitrate (UAN) applied accordingly (variable rate). In addition to the variable rate treatment, a fixed rate and a check plot (no N applied) were evaluated in a randomized complete block experiment. The PNSI was highly correlated with estimates of wheat forage N uptake at all locations and stages of growth. Wheat grain yields increased significantly as a result of applying topdress N in both the fixed rate and variable rate treatments when compared to the check (no topdress N applied). However, no significant differences in wheat grain yield were found when comparing the fixed and variable rate treatments. Variable N rate treated plots (based on PNSI) resulted in a total N savings between 32 and 57 kg N ha–1 when compared to the fixed topdress N rates. In addition to improving site-specific N use efficiency, this technology will likely decrease the risk that overfertilization poses to the environment.

Performance of an electronic yield monitor was evaluated in controlled and field environments. Laboratory experiments were conducted with forces applied by a universal testing machine on the grain flow sensor of the yield monitor. The yield monitor was then field tested on a combine on two different row lengths. The yield reported by the monitor in the laboratory was closely related to a calculated yield (r > 0.99). Yield results reported by the monitor in the field for the 100-m-long strips closely agreed with yields based on the amount of grain as determined by scales. When the monitor was used on 20-m rows there was more scatter in the results because of influence of starting and stopping delays as well as isolated pulses of grain out of the combine’s grain elevator. The goal of splitting the signal from a yield monitor on a continuously moving combine to provide spot yields within a field was not completely accomplished. Field results reported here and by users indicate that yield mapping with continuous flow monitors is feasible. These results further suggest that yield monitors are mechanically reliable and will provide accurate enough information for producer’s yield maps.

An optically–based cotton flow rate measurement device was developed for use in a site–specific yield mapping system. Variations of the final design were field–tested in Tennessee during three harvest seasons. System performance was evaluated by harvesting loads of cotton, integrating flow rate measurements over time to obtain a predicted total mass for each load, and comparing the predicted mass values with actual mass measurements acquired using scales. Integrated flow rate measurements consistently predicted actual load masses with a mean absolute (unsigned) error of 4.0% for all of the loads harvested during the tests. The mean error (signed) was –0.1% with a sample standard deviation of 5.1%. Sensors were never cleaned during any of the harvest seasons, and performance degradation due to debris accumulation on sensor faces was not observed. A subset of test data was analyzed to determine how changing field conditions affected system performance. Data were evaluated for detectable performance effects due to differences in cotton variety, seed cotton moisture content, average cotton flow rate, total load weight, and/or the amount of time elapsed since last system calibration. Cotton variety and average cotton flow rate were shown to affect load mass prediction errors. These results suggest that the system calibration should be checked when the variety being harvested changes, as additional calibrations may be necessary to provide optimum results, and that further optimization of flow rate prediction algorithms may be possible.

Yield mapping has become an important function in site–specific management systems. Decisions of economic significance are made based upon the patterns and summary statistics of the yield data recorded during harvest. Unfortunately, yield maps frequently contain data points that are not accurate estimates of the yield at that point. The yield estimate at any given point is affected by a number of factors, including the number of combines used in a field, the shape of the field, yield and moisture calibration, harvest pattern, and operator practices. By relying on limits and patterns in yield data, filtering can be done prior to mapping to remove some of the problems. Yield data generated by producer cooperators was obtained for use in several research projects. An unsupervised filtering technique for exported yield files was developed and tested on 10 fields of corn, sorghum, and rice. The inaccurate yield points were identified with filter functions based on yield limits, moisture limits, travel distance, yield surges, and less than full header width. This filtering algorithm resulted in a higher field average and lower standard deviation than either the unfiltered data or data filtered with maximum and minimum thresholds alone. The yield data filter removed up to 11% of the data points, with yield distributions being primarily affected at the upper and lower extremes. The filter was judged successful in improving yield map accuracy.

An optically–based system for measuring cotton flow rate was designed and tested in both field and laboratory environments. Accuracy was documented by comparing actual and predicted load weights. When field–tested on a cotton harvester, results were positive, with an average absolute error of 4.7%. A laboratory test was subsequently designed and conducted to gather data describing relationships between system performance and variables such as cotton flow rate, cotton moisture content, and cotton variety. The accuracy results of the laboratory test confirmed those generated in the field, with an average absolute error of 3.4%. Additionally, laboratory test results indicated a moderate correlation between average flow rate and absolute error. Variety also affected system performance, with an average absolute error of 2.4% for one variety, versus 4.9% for another. Moisture content had no detectable effect on accuracy in the laboratory test. The technology described herein has been patented and licensed to industry for application on mobile equipment.

A labor tracking and yield mapping system for manual citrus harvesting was developed and evaluated in two commercial harvesting operations in Florida. The system consisted of button recognition and barcode units, integrated with a differential global positioning system (DGPS) unit mounted on a truck, which was used to handle fruit collection containers (tub or bin) in the citrus grove. When the truck operator stopped to load each full tub or bin, the button and barcode of the harvester who filled the container was activated. This action recorded the number or name assigned to the harvester and the DGPS location of the container. The button unit was more reliable and was designed with an audible (alarm) and a visible (light) feedback to provide the truck operator verification of the recorded information. The system tallied the number of containers each harvester had harvested for the day and allowed the entry of other information pertinent to the harvesting operation. The DGPS data were used to map the yields with ArcView software. Field tests showed the labor tracking system was reliable and user friendly.