Development of Pattern Recognition Techniques for Identification and Counting of Individual Seedlings in Eucalyptus spp Plantations from High Definition Aerial Images
Guilherme Pereira Jorge Franzé1; Emanuel Rocha Woiski2; Douglas Domingues Bueno2; Luiz Carlos Sandoval Góes1; Eder Alves de Moura1
1 Instituto Tecnológico de Aeronáutica, São José dos Campos; 2 Universidade Estadual Paulista
doi:10.20906/CPS/CILAMCE2017-0942
Resumo
Unmanned aerial vehicles (UAVs) for overflight of plantation stands have been used repeatedly by agribusiness, notably the paper and pulp industries. This is an alternative that has been encouraged by the growing supply of these vehicles, as well as changes in legislation that will allow the use of UAVs more widely in airspace. The UAVs enable the acquisition of high definition aerial images. However, like any other data, the images alone represent nothing without the proper use of pattern recognition, reconstruc- tion, and interpretation techniques. Thus, there is a strong demand for the development of procedures and computational algorithms for this purpose, since the results of these analyzes can lead to the development of business intelligence (BI), allowing the decision maker to take the best action in timely manner. With the help of Scientific Python Libraries from the literature, an appropriate methodology was developed in this work for the automatic counting of individual seedlings in plantations of Eucalyptus spp from high definition photographs. Beginning from the basic concepts of image and color spaces, the histogram was defined as an important feature, as well as the fundamental arithmetic and logical operations, from which morphological and spacial filter are built. Moving on, Precision Agriculture (PA) was defined and some generic issues for its adoption are explored. The problem to be investigated was presented and solved with help of HSV colorspace conversion. With the algorithm properly validated on the training data, an actual business case of seedlings detection and counting out of a mosaic aerial image was proposed as testing data. The high- definition pictures were taken by spectral sensor onboard an UAV from an Eucalyptus spp plantation stand of approximately 62 acres (25 hectares) and provided by Eldorado Brasil. The results were considered very encouraging, stimulating future works in this line of research.
Palavras-chave: Eucalyptus spp.; Python; Image processing; UAVs; Binarized classification; Precision agriculture