Multisensor Data Fusion Implementation for a Sensor Based Fertilizer Application System

Abstract

"Mapping systems" (“mapping approach”), real-time sensor-actuator systems ("sensor approach") or the combination of both (“Real-time approach with map overlay”) determine the process control in mobile application systems for spatially variable fertilization. Within the integrated research project “Information Systems Precision Farming Duernast” (IKB Duernast) the implementation of the “Real-time approach with map overlay” was done for intensive nitrogen fertilization. The bottom line of this sophisticated approach is a comprehensive situation assessment, a typical multisensor data fusion task. Based on a functional and procedural modelling of the multisensor data fusion and decision making process, it could be pointed out that an expert system is an adequate fusion paradigm and algorithm. Therefore, a software simulation with an expert system as core element was implemented to fuse on-line sensor technology measurements (REIP), maps (yield, EM38, environmental constraints, draft force) and user inputs in order to derive an application set point in real-time. The development of an expert system can be viewed as a structured transformation in five levels from the “specification level”, the “task level”, the “problem solving level” and the “knowledge base level” to the “tool level”. In the “tool level” the hybrid expert system shell JESS (Java Expert System Shell) was selected for implementation due to the results of preceding levels. Knowledge acquisition was done within another IKB-subproject by the means of data mining. Typical and maximal times of 10 ms and 60 ms for one fusion cycle were measured running this application on a 32-bit processor hardware (Intel Pentium III Mobile, 1 GHz).