Método de reconstrução tomográfica de amostras agrícolas com o emprego de técnicas big data

• Autor(a) principal: Alves, Gabriel Marcelino
• Data de Publicação: 2020
• Tipo de documento: Tese
• Idioma: por
• Título da fonte: Repositório Institucional da UFSCAR
• Texto Completo: https://repositorio.ufscar.br/handle/ufscar/12726
• topic: Reconstrução de imagens tomográficas; Seleção de projeções tomográficas; Processamento de imagens; Agricultura de precisão; Tomographic image reconstruction; Image processing; Precision agriculture; Tomographic projections selection; Big Data; CIENCIAS EXATAS E DA TERRA::CIENCIA DA COMPUTACAO::METODOLOGIA E TECNICAS DA COMPUTACAO; CIENCIAS EXATAS E DA TERRA::CIENCIA DA COMPUTACAO::SISTEMAS DE COMPUTACAO; CIENCIAS EXATAS E DA TERRA::CIENCIA DA COMPUTACAO

A new method of high resolution tomographic reconstruction of agricultural samples is presented, which uses the spectral density of X-ray tomographic projections as a criterion to minimize processing time and obtain good quality digital images, besides being scalable. The use of the spectral density of the tomographic projections made it possible to evaluate the associated energy in each projection and consequently the amount of information that is related to its probabilities. Thus, the tomographic projections were organized into energy classes and those with the most expressive amounts of information were selected. As part of the method, after selecting projections, Filtered Back Projection (FBP) and B-Spline interpolation were considered to obtain 2D and 3D (volumetric) reconstruction, steps that were parallelized considering the Apache Spark environment. For the execution of the developed method was organized a Big Data environment that had a cluster, installed on the Amazon Web Services (AWS) platform and a stack of technologies. The Big Data environment configuration assessment considered four sets of projection matrice of the same plexiglass heterogeneous phantom totalizing 7840 matrice (35.63 GB) which were processed for 12 different configurations totalizing 427.56 GB of processed tomographic data. The cluster configuration was defined after evaluating the Speedup and Efficiency metrics for the method running in the Big Data environment. In addition, a cluster consisting of a heterogeneous plexiglass phantom, a Sheep-Logan phantom and a homogeneous sample plus 33 seed samples was prepared for the purpose of validating and evaluating the quality of cluster reconstruction of selected tomographic projections. In this context, an image dataset containing 66, 642 2D images of seeds (242 GB) has been organized. The Structural Similarity Index (SSIM), Normalized Root Mean Square Error (NRMSE), and Peak Signal-to-Noise Ratio (PSNR) metrics were used in the validation steps. The SSIM metric was calculated for each projection matrix and the median measurement for the SSIM values of each sample was observed. In this sense, the SSIM analysis showed that the tomographic reconstruction of the two-dimensional samples from the selected projections led to the SSIM value exceeding 0.80 for all samples analyzed. The results showed that the method allowed a reduction between 28% and 38% in the number of tomographic projections in each sample analyzed, without compromising the quality of the reconstructed images. Finally, this new method has been shown to be useful for the analysis of large quantities of agricultural samples based on the use of X-ray tomography in order to meet the management based on precision agriculture paradigms, where the increasing number of analyzes required for agricultural samples in the decision-making process is considered a prime factor.