Main content area

An open-source spatial analysis system for embedded systems

Coelho, Andre Luiz de Freitas, de Queiroz, Daniel Marçal, Valente, Domingos Sárvio Magalhães, Pinto, Francisco de Assis de Carvalho
Computers and electronics in agriculture 2018 v.154 pp. 289-295
Internet, algorithms, cluster analysis, computer software, computers, crop yield, data collection, kriging, monitoring, precision agriculture, soil, user interface
Soil and plant monitoring systems are important tools for applying precision agriculture techniques. To acquire soil-plant system data, the user establishes a sampling strategy, goes to the field, collects data and finally goes to the office for data analysis. Sometimes, when the analysis is performed, the user realizes that the sampling strategy was not adequate and needs to return to the field in order to collect more data. To avoid problems with the sampling strategy, the solution is to have a system that performs the data analysis immediately after its collection, while the user is still in the field. To do that, we can use single board computers; these types of platforms have ports to communicate to sensors and good processing capabilities. Therefore, the objective of this work was to develop an embedded system to perform spatial variability data analysis in the field, right after data acquisition. The software was developed using Python 3.6; the PyQt Integrated Development Environment (Riverbank Computer Limited, Dorchester, United Kingdom) was used to design a graphical user interface. The BeagleBone Black board, running Debian version 8.6, was used to implement the software. The analysis was divided into three steps: in the first one, an outlier and inlier analysis was performed to remove unwanted data; in the second one, the semivariogram was generated, and the variable and standard deviation map was produced by performing ordinary kriging; and in the last one, a cluster analysis was performed to create management classes using a fuzzy k-means algorithm. The graphical user interface showed the variable map and the variable classes map. To test the developed software, soybean yield data that was collected in a 31.6-ha field were used. The developed software was shown to be efficient at performing the spatial variability of soybean yields. The comparison of the generated maps shows the importance of filtering the data before performing the analysis. The developed software is available at the GitHub website.