Your search found 1 records
1 Torres, A. B. B.; da Rocha, A. R.; Coelho da Silva, T. L.; de Souza, J. N.; Gondim, R. S. 2020. Multilevel data fusion for the internet of things in smart agriculture. Computers and Electronics in Agriculture, 171:105309. [doi: https://doi.org/10.1016/j.compag.2020.105309]
Decision support systems ; Internet ; Agriculture ; Irrigation ; Soil moisture ; Evapotranspiration ; Energy consumption ; Linear models ; Sensors ; Crops ; Cashews ; Coconuts / Brazil / Paraipaba
(Location: IWMI HQ Call no: e-copy only Record No: H049724)
(7.91 MB)
The Internet of Things (IoT) aims to enable objects to sense, identify, and analyze the world, but to achieve such goal cost-effectively, it should involve low-cost solutions. That implies a series of limitations, such as small battery life, limited storage capabilities, low accuracy, and imprecise sensors. Data fusion is one of the most widely used methods for improving sensor accuracy and providing a more precise decision. Therefore, we propose Hydra, a multilevel data fusion architecture, to improve sensor accuracy, identify application target events, and make more accurate decisions. Hydra is composed of three layers: low-level (sensor data fusion), medium-level (events and decision making), and high-level (decision fusion based on multiple applications). In partnership with Embrapa (Brazilian Agricultural Research Corporation), we instantiated Hydra for the smart agriculture domain, and we also developed two applications aiming smart water management. The first application goal was to determine the need for irrigation based on soil moisture levels, and the second ascertained the adequate irrigation time by estimating the crop’s evapotranspiration (rate of water evaporation by the soil and transpiration by plants). We performed a set of experiments to assess Hydra: (i) evaluation of methods to detect and remove outliers; (ii) analyze data resulting from the applications; (iii) the use of machine learning to create a new accurate evapotranspiration model based on the sensors data. The results indicate that a combination of the ESD method (Extreme Studentized Deviate) and WRKF filter (Weighted Outlier-Robust Kalman Filter) was the best method to identify and remove outliers. Moreover, we generated an evapotranspiration model using the SVM (Support Machine Vector) quadratic machine-learning model that produced values close to the evapotranspiration reference model (Penman-Monteith).
Powered by DB/Text WebPublisher, from
