AVALIAÇÃO DE METODOLOGIAS DE AQUISIÇÃO DE IMAGENS PARA FINS DE ANÁLISE DE QUALIDADE DE SEMENTES DE MILHO

Abstract

Brazil is one of the largest seed producers in the world. Therefore, one must be sought ways to ensure seed quality, free from mechanical damage and/or infection. Computational techniques can assist on detecting these problems, indicating seeds that are or are not following the standards of the existing legislation. The use of digital image processing in agriculture has been established in various areas of knowledge. In agriculture, the digital image processing can help in the area of visual inspection of seeds, a tedious task and with human subjectivity. The digital image processing consists of several steps: image acquisition, preprocessing, segmentation, recognition and interpretation. Among all these steps, the acquisition is the most important, since following stages depend on it to get the desired information. The acquisition has an image capture device, lighting, among other items. An acquisition system that is not developed to a defined purpose can produce inefficient images for the image processing system. For instance, the system can detect a non-existent disease or mechanical damages in the seed, presenting false positives and false negatives, due to the presence of residues in the environment or a wrong image resolution. The objective of this work is to choose the best methodology for capturing images for analysis of seed quality of maize. This was done through two measures of distances between histograms, the intersection and the correlation. To evaluate the performance of the developed methodologies in terms of repeatability and reproducibility, three replications of nine image groups were performed after the systematic rearrangement of equipment in each repetition; and three replicates of nine image groups in which the equipments were not removed from setup; at the end, each repetition had 81 images. As a result, it was verified the need to perform a calibration procedure of the acquisition system at each repetition; and that there is a constancy in the images, for the same repetition, obtaining, for the best case, a distance between histograms of 0.99804 ± 0.00124, with the correlation metric.