Michelle Chaves Kuroda

A fast approach for unsupervised karst feature identification using GPU

Abstract Among the geological features, karst is the one that has received special attention in oil and gas exploration for being a strong indicator of the potential existence of hydrocarbon reservoirs. The integration of automatic pattern recognition methods and Graphics Processing Units (GPU) provides a powerful tool to help geological interpretation of seismic data. In order to provide insightful information for interpreters, this work investigates the usage of GPUs in addition to image segmentation by means of unsupervised classification for the identification of karst features in 3D seismic data. For this purpose, an implementation of the robust Self-Organizing Map for GPUs (SOM/GPU) is provided, and a comparison against a Central Processing Unit (CPU)-based SOM (SOM/CPU) is performed to assess the speeding-up provided by GPU. Experiments have shown promising results for geological interpretation using seismic data.

Learning to Classify Seismic Images with Deep Optimum-Path Forest

Due to the lack of labeled information, clustering techniques have been paramount in the last years once more. In this paper, inspired by the deep learning phenomenon, we presented a multi-scale approach to obtain more refined cluster representations of the Optimum-Path Forest (OPF) classifier, which has obtained promising results in a number of works in the literature. Here, we propose to fill a gap in OPF-based works by using a deep-driven representation of the feature space. Additionally, we validated the work in the context of high resolution seismic images aiming at petroleum exploration, as well as in general-purpose applications. Quantitative and qualitative analysis are conducted in order to assess the robustness of the proposed approach.

Analysis of porosity, stratigraphy, and structural delineation of a Brazilian carbonate field by machine learning techniques: A case study

AbstractThe upscaling of well logs has many challenges, especially for carbonate rocks. Primary among them is the suitable choice of seismic attributes to be integrated with well information, whose random combination can produce artifacts of rock properties. To solve this problem, we have developed an alternative hybrid method to estimate well-log data from seismic attributes, associating the seismic attributes choices with the genetic algorithm and artificial neural network multilayer perceptron ability to predict neutron porosity. Thirty-seven seismic attributes were extracted along 12 wellbores from an Albian offshore carbonate reservoir of the Campos Basin. From these attributes, three were selected: 3D mix, structure-oriented median-filtered amplitude, and acoustic impedance. From this set of seismic data, we used the first two attributes to estimate neutron porosity in the reservoir seismic area. As a result, we obtained a 3D map of well-log information at the seismic scale. In the 3D map, it is pos...

Self-organizing maps as a tool for segmentation of Magnetic Resonance Imaging (MRI) of relapsing-remitting multiple sclerosis

Multiple Sclerosis (MS) is the most prevalent demyelinating disease of the Central Nervous System, being the Relapsing-Remitting (RRMS) its most common subtype. We explored here the viability of use of Self Organizing Maps (SOM) to perform automatic segmentation of MS lesions apart from CNS normal tissue. SOM were able, in most cases, to successfully segment MRIs of patients with RRMS, with the correct separation of normal versus pathological tissue especially in supratentorial acquisitions, although it could not differentiate older from newer lesions.

Oil reservoir quality assisted by machine learning and evolutionary computation

The main target of oil and gas exploration companies is to identify reservoirs and their location with high accuracy. For this purpose, all efforts are applied to reduce uncertainties and risks of water contamination or drilling of dry wells in order to extract as much as possible from the subsurface in the shortest time and at the lowest cost. This chapter shows an alternative for the combination of machine learning techniques, evolutionary computation, and geological interpretations to decrease uncertainties in identifying the location of favorable reservoirs. For this purpose, seismic and well log data from a sand Brazilian field were analyzed. The identification of sandy facies as conducers was made by means of self-organizing maps and extrapolated into signals of seismic data by probabilistic neural networks, converting the image of original amplitude into rock properties. The genetic algorithm was also tested to evaluate different seismic attributes among a group of 37 possibilities to perform the facies prediction task. The image description by multiattributes allowed the definition of the facies distribution modeling. The same process was applied to predict the probability of porosity distribution in seismic data by multilayer perceptron and generalized regression, once again using the genetic algorithm. Through these properties, models from two favorable areas of reservoir were identified in the southwest part of the field. Core description corroborates with the results found by the suggested methodology, indicating its satisfactory application.