Pengembangan metode pengendalian gulma pada pertanian presisi berbasis multi agen komputasional

Abstract

Weed control practices commonly performed on crop farming are preemergence during land preparation and after planting. Weed control in land preparation activities are carried out with a view to facilitate the work of machinery and equipment, while post emerge applications done with a view to ensure main crops can survive in the critical period which the competition of main crop and weeds tend to be very tight. Machine vision system allows monitoring of plant diseases or nutrient deficiencies for proper treatment. Selection of weed control scenarios in agricultural land varies considerably depending on several factors. Failure of weeds control is not only caused by errors in identifying weed species, density and selection of herbicides attack but also weather conditions such as wind, temperature, relative humidity and rainfall affect the effectiveness of spraying applications and potential waste by run-off and drift. Supervisory system allows users to select the type of technology and the best weed control scenario. The purpose of this research are : 1) to design supervisory system to determine the type of technology and rate of the herbicide applicator and spray the control precision farming practices based on multi-agent systems and 2) to design the method to identify main plants and weeds, and weeds attack density analysis using the parameters contained in the image. The supervisory system design consists of user preference and dialogue, knowledge module and an inference engine. Knowledge modul contains a collection of several modules related to weed control including the basis of intelligent systems, technologies database, and parallel computing base. Multi intelligent agents on the basis of intelligent system modules were involved in weed control activities in accordance with the needs on the field. Forward velocity target is achieved by operating multiple processors in parallel where each processor is dedicated to perform a particular job through an intelligent agents. Location accuracy and spraying dosage accuracy are designed by dividing the image into several segments in accordance with the size of the applicator width. Image partition method gives an application map that will guide the equipment to distribute the herbicide more precisely in therm of dose and location. In a pre-emerge application during land preparation, weed control system with 4 agents showed speed up values of 3.775 on 100 000 to unlimited the number of jobs, with an efficiency of 93.88%. While on post emerge application, weed control system with 4 agents shows the value of speed up reaches 3.814 on 100 000 to unlimited the number of jobs with an efficiency of 95.35% .