The Use of Computational Intelligence for Precision Spraying of Plant Protection Products

• Main Author: Faiçal, Bruno Squizato
• Publication Date: 2016
• Format: Doctoral thesis
• Language: eng
• Source: Biblioteca Digital de Teses e Dissertações da USP
• Download full: http://www.teses.usp.br/teses/disponiveis/55/55134/tde-02032017-155603/
• topic: Agricultura de precisão; Agricultural spraying; Deposition prediction; Precision agriculture; Predição da deposição; Pulverização agrícola

Protection management with the aid of plant protection products makes it possible to carry out pest control programs in agricultural environments and make them less hazardous for the cultivation of products on a large scale. However, when these programs are put into effect, only a small proportion of the sprayed products is really deposited on the target area while much of it is carried to neighboring regions. The scientific literature includes studies on the use of mathematical techniques to calculate the physical transformation and movement and provide a deposition estimate of the product. On the basis of this prediction, it is possible to configure a system which can allow the spraying to be carried out in normal weather conditions in the region for a satisfactory performance, although these conditions can undergo changes and make any statistical configuration unreliable. An alternative way of overcoming this problem, is to adapt the spray elements to the meteorological conditions while the protection management is being undertaken. However, the current techniques are operationally expensive in computational terms, which makes them unsuitable for situations where a short operational time is required. This thesis can be characterized as descriptive and seeks to allow deposition predictions to be made in a rapid and precise way. Thus it is hoped that the new approaches can enable the spray element to be adapted to the weather conditions while the protection management is being carried out. The study begins by attempting to reduce costs through a computational model of the environment that can speed up its execution. Subsequently, this computational model is used for predicting the rate of deposition as a fitness function in meta-heuristic algorithms and ensure that the mechanical behavior of the spray element can be adapted to the weather conditions while the management is put into effect. The results of this approach show that it can be adapted to environments with low variability. At the same time, it has a poor performance in environments with a high variability of weather conditions. A second approach is investigated and analyzed for this scenario, where the adaptation requires a reduced execution time. In this second approach, a trained machine learning technique is employed together with the results obtained from the first approach in different scenarios. These results show that this approach allows the spray element to be adapted in a way that is compatible with what was provided by the previous approach in less space of time.