'Smart Sprayer' Expert System for Site-specific Weed Management
L. F. Tian,* J. F. Reid
Illinois Council on Food and Agricultural Research

The objective of this project is to capitalize on the new 'smart sprayer' technology to more effectively reduce the herbicide application amount and prevent agricultural pollution from the source. Because of the novelty of this technology, no database information is available. This project is going to establish such an information base. Detailed diagnostics are expected to reveal how the system design effects the overall performance so that the technology can be further optimized and brought closer to commercialization. We will improve the computer-vision guided smart sprayer system to create a local (field) environment-aware applicator as the research platform.

Development of Precision Herbicide Application System
L. F. Tian,* J. W. Hummel
University of Illinois

The goal of this project is to develop a precision herbicide application (robotic) system for low-input pest control strategies in soybean and maize production. Specific objectives include (1) evaluating the agronomic and environmental benefits of low-input herbicide applications and the status of current technology in this area, (2) developing practical real-time prototype systems for individual plant sensing and equipment control, (3) conducting on-farm trials to evaluate the prototype under the constraints of normal farm operation. With this precision system, herbicide would be only applied to the target plants in the fields.

Improved Application of Pest Control Substances
L. F. Tian*
University of Illinois; U.S. Department of Agriculture

Equipment and techniques are being developed to improve the application efficiency of agricultural chemicals. Droplet size spectra from various atomizers are measured to determine target coverage versus spray drift potential. Field studies of spray drift deposits are used to verify the droplet size evaluations. Sensors and automatic control systems are being developed to apply pest control substances as a function of soil organic matter, travel speed, and other input variables. Techniques for incorporation of herbicides in the soil profile of conservation tillage systems are being developed and evaluated.

Large-Scale Remote Sensing for Precision Farming
L. F. Tian*
University of Illinois

The research is concerned with innovation and further technological development of living biological object high-resolution remote sensing techniques. The project includes: (1) the sensor calibration and outdoor image formation technologies and (2) the knowledge-based image-processing algorithms. The first method is based on the concept that the growth conditions are reflected in both the spectral and morphological features of the biological objects. The second part incorporates a priori chronological constraint into biological object image interpretation procedure. It promises to provide a unique capacity to reveal the living biological object's growth condition with higher efficiency and accuracy.

Low-Input and Nonchemical Weed Control System
L. F. Tian,* J. F. Reid
University of Illinois

In an effort to reduce herbicide application amounts, this research will integrate a machine vision-sensing system with a herbicide sprayer system to create an intelligent sensing and mapping system for pest control. We will first develop a real-time plant-sensing and spraying system based on field experiments to characterize the plant features that can be used for crop detection. The integrated system will be field tested under varying field conditions. Outcomes of this research will provide a system for increasing farm sustainability and protecting water quality through precision application of herbicides. The vision technology will provide a precision sensor that can be further developed for precision mapping of field weed infestations and vehicle guidance.

Site-Specific Crop Management
C. E. Goering,* J. Liu, L. F. Tian, R. Hornbaker, D. Bullock
University of Illinois

Setting realistic, spatially referenced yield targets is a key step in site-specific crop management. Using records from 30 years of corn production on the Morrow Plots, an artificial neural network (ANN) was trained to calculate corn yields as a function of the soil, weather, and management factors influencing such yields. The trained ANN predicted corn yields with an rms error of approximately 20%. A genetic algorithm was used with the trained ANN to determine the combination of input factors producing maximum yield, which was about 75% greater than the maximum observed yield. The next step is to study whether the ANN can be retrained for fields with much smaller databases.