Won Suk Lee
Biological & Ag. Engineering, UC Davis
A weed can be thought of as any plant growing in the wrong place at the wrong time and doing more harm than good. Weeds compete with the crop for water, light, nutrients and space, and therefore reduce crop yields and also affect the efficient use of machinery (Parish, 1990). Many methods are used for weed control. Among them, mechanical cultivation is commonly practiced in many vegetable crops to remove weeds, aerate soil, and improve irrigation efficiency, but this technique cannot selectively remove weeds located in the seedline between crop plants. The most widely used method for weed control is to use agricultural chemicals (herbicides and fertilizer products). In fact, the success of U.S. agriculture is attributable to the effective use of chemicals. For example, the total of 5.9 million kg of agricultural chemicals (herbicides, insecticides, fungicides, and other chemicals were used to produce processing tomatoes in California alone in 1994 (USDA, NASS and ERS, 1995). This heavy reliance on chemicals raises many environmental and economic concerns, causing many farmers to seek alternatives for weed control in order to reduce chemical use in farming. For some crop/weed situations there are no selective herbicides. Selective herbicides selectively kill only weeds and not crop plants, thus they are important for weed control.
Since hand labor is costly, an automated weed control system may be economically feasible. A real-time precision robotic weed control system could also reduce or eliminate the need for chemicals. Although there have been many efforts to control in-row weeds, no system is currently available for real-time field use. In this research, an intelligent real-time robotic weed control system has been developed to identify and locate outdoor plants for selective spraying of in-row weeds using an environmentally sound and friendly chemical application system based upon machine vision technology, pattern recognition techniques, knowledge-based decision theory, and robotics.
Tomatoes are one of the leading vegetable crops produced in California. In 1996, over 9 billion kg of processing tomatoes were produced in California, accounting for 93% of all processing tomatoes produced in the U.S. (USDA and NASS, 1997). However, the current in-row weed control method is highly dependent on labor-intensive and costly hand hoeing. A significant amount of manual work is still required for weed control in crop rows, which hopefully can be automated with today's rapidly growing state of the art computer technologies.
Ever since humans started farming, weeds have been one of the major obstacles to maximizing production. Until recent technologies evolved, farming has been dependent on human power. Although there have been a lot of efforts to control in-row weeds effectively, no system is currently available to replace tedious and time-consuming hand weeding.
Hand hoeing is costly, time consuming and labor intensive. For example, the cost for hand weeding was about $80 per 0.4 ha (1 acre) for processing tomato production in northern California in 1996. In California, there were over 12 million ha of farm land in 82,000 farms in 1996. Among them, 140,183 ha were used for tomato production.
There has been a lot of effort to control weeds non-chemically in order to reduce chemical costs in response to environmental pressure. These methods can be largely divided into cultural weed control methods, mechanical control methods, and biological control methods. In this research, mechanical control methods were the main focus. These include hand pulling or hoeing, tillage, cultivation, burning, flame cultivation, and electrical devices (Cooperative Extension Service (1995)). There were some researchers who investigated non-chemical weed control methods (Parish (1990) and Bond (1992)), but as Bond pointed out, few attempts have been made to selectively control weeds in the seedline. However, with advances in image processing and machine vision technologies, many researchers have applied these techniques to agriculture to identify individual crop plants.
Image processing and machine vision technologies have been applied successfully to many agriculutral settings recently. Properly applied machine vision techniques improve manufactured product quality and provide valuable process control information (Novini, 1992). Their primary agricultural applications are automatic inspection and sorting of agricultural products (Shearer and Payne (1990), Miller and Delwiche (1991), Al-Janobi and Kranzler (1994), Lan et al. (1996) and Crowe and Delwiche (1996a, b)), and identifying and locating individual crop plants (Jia et al. (1990) and Tian and Slaughter (1993)). Machine vision has been also used for guiding a robotic system in the harvest of fruits (Slaughter and Harrell (1989)), detecting fruit defects, evaluating chemical applications (Jiang and Derksen (1993)) and investigating plant architectural measurments (Tarbell and Reid (1989)).
The goal of this project was to build a real-time machine vision robotic weed control system that can detect crop and weed locations, kill weeds and thin crop plants. The system needed to recognize tomato plants and weeds outdoors in commercial tomato fields using image processing techniques while moving forward at a constant speed. Specific objectives were to:
The objective of this research was to develop a real-time robotic weed control system for seedline weds and to replace costly hand weeding with robotic weed control using machine vision, pattern recognition, and robotics. All research was conducted with juvenile processing tomato plants in commercial tomato fields in northern California during 1996 and 1997. Tomato images were acquired in 5 different fields in 1996 and in 8 different fields in 1997.
The UC Davis Robotic Cultivator (Slaughter et al., 1997) was utilized as a guidance system to center the prototype system over the row. A seedline image was acquired for the recognition of plant leaves. Plant leaves were identified using their shape characteristics and the Bayesian discriminant function. After distinguishing tomato plants from weeds, the main computer sent the weed locations to a precision chemical application system which opened a corresponding spray valve in the valve/nozzle array when it was above a weed plant. The precision chemical application system consisted of 8 valves/nozzles, which due to space constraints were aligned in two rows (four in each row). For precise herbicide application, the image was subdivided into a spray grid of 8 rows by 18 columns. Each cell in the spray grid corresponded to a 1.27 cm by 0.64 cm region on the seedbed. The precision spray system was capable of applying chemical herbicides to individual spray cells providing a level of precision unparalled in existing agricultural spray systems.
A real-time intelligent robotic weed control system was developed and tested for non-occluded plant leaves in commercial processing tomato fields for selective spraying of in-row weeds using a machine vision system and a precision chemical application system. The machine vision system was composed of the SHARP image processing boards, a color video camera, a multifunction I/O board, a Pentium Pro 200 MHz CPU, and plant recognition algorithms. The precision chemical application system was composed of a microcontroller, a manifold, a specially designed accumulator, 8 solenoid valves and micro-spray nozzles, valve control circuits, a valve control software. A Bayesian decision rule and a real-time LUT conversion board (AUXLUT card) were utilized in segmenting color images and thus made real-time implementation possible.
The image processing algorithm took 0.344 s to identify 10 tomato cotyledons in the image using only the features of ELG and CMP, for one frame of a 256 by 240 pixel image representing a 11.43 cm x 10.16 cm field of view. Thus, the prototype cultivator could travel at a continuous rate of 1.20 km/h.
The prototype robotic weed control system was tested in commercial tomato fields in Northern California from March to May 1997. Overall the robotic weed control system correctly identified and did not spray 75.8% of the tomato plants and correctly sprayed 47.6% of the weeds.
last modified 12/17/2003