Automatic Detection of “Palmeira da Macaúba”(Acrocomia Aculeata) in digital images using Mathematical Morphology
ISBN 978-85-88783-11-9
Authors
1Lopes Braga Fonseca, A.; 2Miranda Nunes, D.; 3Fonseca Neto, F.; 4de Assis de Carvalho Pinto, F.; 5Gripp Junior, J.; 6Botelho, M.F.
1UFV Email: alebragaifes@gmail.com
2UFV Email: darlan.m.nunes@gmail.com
3UFV Email: franciscogeoifes@gmail.com
4UFV Email: facpinto@ufv.br
5UFV Email: jgripp@ufv .br
6IFSMG Email: mfbotelho@yahoo.com.br
Abstract
The increase of the world economy in the last decade resulted in considerable increase in fuel consumption. In this context, the biodiesel production becomes a sustainable alternative in environmental, social, and economic point of view. In Brazil, several species are considered potential immediate economic exploitation, among which stands out the palm of macaúba, due to the high productivity of oil macaúba, besides being the Palm of greater dispersion in Brazil. In this context, the present study aimed at using techniques Digital Image Processing, in particular the theory of Mathematical Morphology for automatic detection and extraction of palm trees Macaúba (Acrocomia Aculeata) from multispectral aerial images, as an alternative that allows their mapping. This study was developed using aerial imagery obtained from an experimental field of the Federal University of Viçosa (UFV) located in the district of Araponga, Minas Gerais, Brazil. The image was captured for analysis by a camera equipped with a SONY DSC-H5, with a spatial resolution of 0.08 meters and 8-bit radiometric aircraft and comprising bands of blue, green and red. The methodology was divided into two stages: Pre-processing of the images and the implementation of an algorithm for detection and extraction of palm tree macaúba. In the first stage, the pre-processing of the image was carried out, in order to highlight the elements of interest, proceeding the conversion of the image to grayscale, followed by contrast enhancement by histogram equalizatio. For the detection and extraction of palm tree macaúba, a Filter Alter-nating Sequential Reconstruction (FASR) of the open-close type was implemented in MATLAB software in 2013. These filters can be categorized in two ways, according to the order of execution of operators: by closing or opening-closing-opening; in the following ways: Traditional Alternating Sequential Filtering (ASF) or reconstruction (FSAR). Opening and closing by reconstruction enables to remove local intensity peaks of gray without changing the borders of the regions, unlike the traditional opening and closing operations. Opening and closing by reconstruction possible to remove local intensity peaks of gray without changing the borders of the regions, unlike the traditional opening and closing operations. A new image was generated with the filtering FSAR, obtaining a more significant separation of objects image background. The structuring element used was the disc type and a total of five iterations, that is, the structuring element FSAR is incremented by the size of five pixels, allowing filtering image noise while preserving the form of the elements of interest (Palm). For individualization of palm trees and reduction of the image complexity, the thresholding was carried out using Otsu method, whereby a global threshold that seeks to maximize the variance between two classes, through the object pixels and background pixels in the image. We found some limitations in this research, mainly related to the presence of shadows in the image and trees in the plantation areas of the palms, and especially due to the growth stage, they were in the palm of macaúba because many of these were close and, as a result, the extraction returned connected features. By means of a counter implemented by the extraction algorithm, it was possible to compute individualized Palm extracted correctly preserving its form. This individualization provided the low percentage underestimated by the algorithm, compared to the value obtained by the identification method and visual counting. In general, the method of automatic detection of the macaúba palm obtained satisfactory results. As a perspective for future work, in order to overcome the problems associated with the addition of palm trees too close, by region segmentation techniques should be employed, enabling better efficiency in individualization.