Randal K. Taylor

RED EDGE AS A POTENTIAL INDEX FOR DETECTING DIFFERENCES IN PLANT NITROGEN STATUS IN WINTER WHEAT

Under certain conditions normalized difference vegetative index (NDVI) has low sensitivity; therefore red-edge position (REP) has been tested as an alternative vegetative index. The objective of this study was to determine if REP could be useful for detecting differences in N status for winter wheat compared to NDVI. A spectrometer, and the SPAD meter were used to measure N status. Sensitivity to plant N response and different growth stages was found for NDVI and REP, but NDVI sensitivity tended to decrease as N rate increased and REP sensitivity tended to increase with increased N rate and advancing growth stage. Both NDVI and REP were linearly correlated at all growth stages (r2 = 0.85). REP and SPAD meter readings were highly correlated (r2 = 0.62) as were NDVI and SPAD (r2 = 0.56). Overall, REP and NDVI sensitivity at high plant biomass were similar for winter wheat.

Effective Spatial Resolution for Weed Detection

Patch spraying herbicide to control weeds has distinct advantages. Machine vision using digital images can be used for patch spraying however, research has found varied results. Image processing requires considerable computational time when image resolutions are high and poses difficulties for real time application. This study was undertaken to study trade offs between image resolution and detection accuracy. The images were acquired using a digital color camera and then individual plant images of 128x128 pixels were extracted using Matlab software. Shape, green color and radial spectral energy features were selected to classify crop weed images and three resolutions were taken. Excess green method was used for segmentation. A Bayesian classifier was used for three classifications while using only shape features, only green color and spectral energy features, and all features combined. Green color and spectral energy features performed best. The classification accuracy using these features at different resolutions for weed detection varied from 80% to 87%. Their performance was unaffected by image resolution and shows potential for field applications.

Adaboost and Support Vector Machine Classifiers for Automatic Weed Control: Canola and Wheat

Detection of plant species for automatic weed control is the most challenging task. Detection accuracies can be improved by better sensing equipment and by accurate classifiers. Adaboost and support vector machine are two state-of-the-art classifiers. However, there are few studies applying them for plant species discrimination. This study attempted to investigate adaboost algorithms and support vector machine kernels for automatic weed control of canola and wheat. Individual images (64x64 pixels) of crop or weed plants, extracted from field digital color images, were used to extract features for classification. For canola, Real Adaboost was 3.29% more accurate than the Bayesian classifier and for wheat 3.09%. For canola, the radial basis function kernel based support vector machine was 2.67% more accurate than the Bayesian classifier and for wheat 4.77%. Overall, the Real Adaboost algorithm performed best among the selected adaboost algorithms and radial basis kernel among the selected support vector machine kernels.

Nozzles for Variable Rate Fertilizer Application

Variable rate application of liquid fertilizer is challenging with standard fixed orifice nozzles because of the limited rate changes. Variable orifice nozzles have been developed that allow greater rate changes without huge pressure changes. Commercially available variable orifice nozzles were evaluated for variable rate fertilizer application. Eight nozzles from two manufacturers were tested at a range of flows from 0.2 to 0.8 gpm. Nozzles were evaluated for consistency among nozzles and repeatability of individual nozzles. Pressure was measured at each flow and pressure/flow curves were generated. TDVR-03 and VeriTarget showed inconsistent behavior among nozzles at different pressures. When compared to manufactures’ flow data it was noticed that VeriTarget nozzles appear to behave differently. Even though the flow rates of VeriTarget nozzles were different from manufacturers’ rated flow rate, it was observed that all the nozzles showed repeatable results and performed best above 40 psi.

Canola – Weed Identification for Machine Vision based Patch Spraying

Canola oil is healthy cooking oil due to low saturated fat content. Current US demand is met by imports equivalent of 1.2 M ha of production each year. Another emerging use of canola oil is as bio-diesel feedstock. Both demands led to increased acreage and research focus. Early stage weed control is a critical operation to minimize yield losses. This study was aimed to develop machine vision based algorithm for patch spraying to control weeds. Field images of production canola at 4-6 leaves stage were taken with a digital color camera. Individual images (128x128 pixels) of canola or weed were then extracted from field images. Seventy five images were used to train the algorithm and another set of seventy five images was used for testing. Three resolutions levels (100%, 50% and 25%) were used to study effect of resolution on classification. Six shape features, one Fourier, and two green color features were used to classify the images. The 25% resolution level performed poorly for canola identification but was good for weed identification. Overall, the 50% resolution level was best for both plant species. The highest classification accuracies obtained were 96% and 100% for canola and weed plants, respectively for 50% resolution level. The improved classification accuracy at the reduced resolution level (50%) has the potential to significantly affect future machine vision based herbicide applicators.

Machinery Productivity Estimates from Seed Tenders

Several methods and machines have been introduced during the last five years that can improve the timeliness and productivity of planting operations. Several manufacturers claim these devices can increase productivity by more than 50% over conventional methods. This presentation provides a discussion and insights on the improvement of corn and soybean planting systems, while using a seed tender and other similar devices. A comparison between machine operations is analyzed with the assumptions made by these claims. While the claims may be valid, farm clientele deserve to know the conditions under which these improvements can be expected. The results can assist farmers in evaluating how these purchases influence machine productivity, and how to identify potential operational areas that can improve their productivity with existing machinery systems. It also provides better estimates for parameters currently listed as ranges within the Agricultural Machinery Management Data (ASABE D497.5).

On-the-go Sensor System for Cotton Management for Application of Growth Regulators

Sensors can be used to measure reflectance indices for different crops which in turn can be used to estimate crop parameters. Normalized Difference Vegetative Index (NDVI) is one of the common indices. A sensor system was developed to record NDVI data for cotton along with crop height using ultrasonic sensors, to estimate cotton physiological parameters and canopy coverage. The experiments were conducted at Altus, OK. Average values of NDVI and plant height for each plot measured by on-the-go sensors were compared with the manually measured cotton structural parameters. A good correlation for accumulative data of different growth stages was observed between manually measured height (Hm) and height measured by the sensors (R2 = 0.80), showing that ultrasonic sensors can be used to measure height of cotton . For mid July data when there is need of growth regulators, NDVI over the row (NDVIOR) and weighted average of NDVI were significantly correlated with Hm and height to node ratio (HNR). Also it was found that HNR and Hm can be represented as a function of NDVI and plant height, which could be used to construct a real time plant growth regulator (PGR) applicator in future. Strong correlation (R2=0.71) was obtained between Sensor height and NDVIOR which implies that the NDVI sensors can solve the purpose of height sensor in future studies on cotton management. On the other hand HNR was found to be correlated with Hs for one of the studies, which infers that HNR can directly be calculated from crop height for application of PGR for upland cotton.

Evaluation of Pneumatic Granular Herbicide Applicators for Seeding Small Grains in Oklahoma

TWO small central metering, air-delivery granular herbicide applicators were evaluated for potential use in seeding. Soybeans, mungbeans, milo and wheat were metered over a typical range of rates with a Gandy 5812 and a Flex-King 160. Some loss of accuracy during operation with seeder platform at an angle was observed for the Flex-King 160, otherwise metering and uniformity was good for both machines. Some problem with seed damage was encountered with the Gandy 5812 when seeding mungbeans and milo.

System and Algorithm Development of Automatic Corn Plant Identification Using Laser Line-scan Technique

Identifying corn plant location and/or spacing is important for predicting yield potential and making decisions for in-season nitrogen application rate. In this study, an automatic corn stalk identification system based on a laser line-scan technique was developed to measure stalk locations during corn mid growth stages. A laser line-scan technique is advantageous in this application because the line-scan data sets taken from various points of view of a plant stalk results in less interference and higher probability of plant recognition. Data were collected for two 10 m corn rows at growth stage V8 and V10 using a mobile test platform in 2011. Each potential stalk cluster was identified in a scan and registered with the same stalks in previous scans. The final location of a stalk was the average of the measured locations in all scans. The current system setup with data processing algorithms achieved 24.0% and 10.0% of mean total errors in plant counting at V8 and V10 growth stages, respectively. The root mean squared error (RMSE) between system measured plant locations and manually measured ones were 2.3 cm and 2.6 cm at V8 and V10 growth stages, respectively. The interplant spacing measured by the developed system had a good correlation with the manual measurement with an R2 of 0.962 and 0.951 for V8 and V10 growth stages respectively. This system can be ultimately integrated in a variable-rate-spraying system to improve real-time, high spatial resolution variable rate nitrogen applications.