Fault enhancement using probabilistic neural networks and Laplacian of a Gaussian filter: a case study in the Great South Basin, New Zealand.

Fault identification is critical in defining the structural framework for both exploration and reservoir characterization studies. Interpreters routinely use edge-sensitive attributes such as coherence to accelerate the manual picking process, where the actual choice of a particular edge-sensitive attribute varies with the seismic data quality and with the reflectivity response of the faulted geologic formations. CMY color blending provides an effective way to combine the information content of two or three edge-sensitive attributes when more than one attribute is sensitive to faults. We evaluate whether combining the information content of more than three attributes using probabilistic neural networks (PNN) provides any additional uplift. We employ a training data consisting of manually picked faults on a coarse grid of 3D seismic lines, and then we employ an exhaustive search PNN to identify the optimal set of attributes to create a fault probability volume for a 3D survey acquired over the Great South Basin, New Zealand. We construct a suite of candidate attributes using our understanding of the attribute response to faults seen in the data and examples extracted from the published literature to use the list as the analyzed attributes. Using a subset of picked faults as training data, we evaluate which suite of attributes and hyperparameters exhibit the highest validation on the remaining training data. When used together, we find that volume aberrancy magnitude, GLCM homogeneity, GLCM entropy, Sobel filter similarity, and envelope best predict the faults for this dataset. The PNN supervised classification creates a seismic image volume that exhibits fault probabilities providing a simple combination of multiple seismic attributes. We also find that applying a directional Laplacian of a Gaussian and skeletonization filters to the PNN fault volumes provides a superior result to simple CMY blending techniques.

OPTIMAL CONSTRUCTION OF THE PATTERN MATRIX FOR PROBABILISTIC NEURAL NETWORKS IN TECHNICAL DIAGNOSTICS BASED ON EXPERT ESTIMATIONS

In the field of technical diagnostics, many tasks are solved by using automated classification. For this, such classifiers like probabilistic neural networks fit best owing to their simplicity. To obtain a probabilistic neural network pattern matrix for technical diagnostics, expert estimations or measurements are commonly involved. The pattern matrix can be deduced straightforwardly by just averaging over those estimations. However, averages are not always the best way to process expert estimations. The goal is to suggest a method of optimally deducing the pattern matrix for technical diagnostics based on expert estimations. The main criterion of the optimality is maximization of the performance, in which the subcriterion of maximization of the operation speed is included. First of all, the maximal width of the pattern matrix is determined. The width does not exceed the number of experts. Then, for every state of an object, the expert estimations are clustered. The clustering can be done by using the k-means method or similar. The centroids of these clusters successively form the pattern matrix. The optimal number of clusters determines the probabilistic neural network optimality by its performance maximization. In general, most results of the error rate percentage of probabilistic neural networks appear to be near-exponentially decreasing as the number of clustered expert estimations is increased. Therefore, if the optimal number of clusters defines a too “wide” pattern matrix whose operation speed is intolerably slow, the performance maximization implies a tradeoff between the error rate percentage minimum and maximally tolerable slowness in the probabilistic neural network operation speed. The optimal number of clusters is found at an asymptotically minimal error rate percentage, or at an acceptable error rate percentage which corresponds to maximally tolerable slowness in operation speed. The optimality is practically referred to the simultaneous acceptability of error rate and operation speed.

STUDY ON THE PARAMETERIZATION RESPONSE OF PROBABILISTIC NEURAL NETWORKS FOR SEISMIC FACIES CLASSIFICATION IN THE GULF OF MEXICO

Nowadays, there are many unsupervised and supervised machine learning techniques available for performing seismic facies classification. However, those classification methods either demand high computational costs or do not provide an accurate measure of confidence. Probabilistic neural networks (PNNs) overcome these limitations and have demonstrated their superiority among other algorithms. PNNs have been extensively applied for some prediction tasks, but not well studied regarding the prediction of seismic facies volumes using seismic attributes. We explore the capability of the PNN algorithm when classifying large- and small-scale seismic facies. Additionally, we evaluate the impact of user-chosen parameters on the final classification volumes. After performing seven tests, each with a parameter variation, we assess the impact of the parameter change on the resultant classification volumes. We show that the processing task can have a significant impact on the classification volumes, but also how the most geologically complex areas are the most challenging for the algorithm. Moreover, we demonstrate that even if the PNN technique is performing and producing considerably accurate results, it is possible to overcome those limitations and significantly improve the final classification volumes by including the geological insight provided by the geoscientist. We conclude by proposing a new workflow that can guide future geoscientists interested in applying PNNs, to obtain better seismic facies classification volumes by considering some initial steps and advice.

Dolomitization geometry and reservoir quality from supervised Bayesian classification and probabilistic neural networks: Midland Basin Leonardian Wichita and Clear Fork Formations

Extensive dolomitization is prevalent in the platform and periplatform carbonates in the Lower-Middle Permian strata in the Midland and greater Permian Basin. Early workers have found that the platform and shelf-top carbonates were dolomitized, whereas slope and basinal carbonates remained calcitic, proposing a reflux dolomitization model as the possible diagenetic mechanism. More importantly, they underline that this dolomitization pattern controls the porosity and forms an updip seal. These studies are predominately conducted using well logs, cores, and outcrop analogs, and although exhibiting high resolution vertically, such determinations are laterally sparse. We have used supervised Bayesian classification and probabilistic neural networks (PNN) on a 3D seismic volume to create an estimation of the most probable distribution of dolomite and limestone within a subsurface 3D volume petrophysically constrained. Combining this lithologic information with porosity, we then illuminate the diagenetic effects on a seismic scale. We started our workflow by deriving lithology classifications from well-log crossplots of neutron porosity and acoustic impedance to determine the a priori proportions of the lithology and the probability density functions calculation for each lithology type. Then, we applied these probability distributions and a priori proportions to 3D seismic volumes of the acoustic impedance and predicted neutron porosity volume to create a lithology volume and probability volumes for each lithology type. The acoustic impedance volume was obtained by model-based poststack inversion, and the neutron porosity volume was obtained by the PNN. Our results best supported a regional reflux dolomitization model, in which the porosity is increasing from shelf to slope while the dolomitization is decreasing, but with sea-level forcing. With this study, we determined that diagenesis and the corresponding reservoir quality in these platforms and periplatform strata can be directly imaged and mapped on a seismic scale by quantitative seismic interpretation and supervised classification methods.

Classification of Metastatic and Non-Metastatic Thoracic Lymph Nodes in Lung Cancer Patients Based on Dielectric Properties Using Adaptive Probabilistic Neural Networks

ObjectiveDielectric properties can be used in normal and malignant tissue identification, which requires an effective classifier because of the high throughput nature of the data. With easy training and fast convergence, probabilistic neural networks (PNNs) are widely applied in pattern classification problems. This study aims to propose a classifier to identify metastatic and non-metastatic thoracic lymph nodes in lung cancer patients based on dielectric properties.MethodsThe dielectric properties (permittivity and conductivity) of lymph nodes were measured using an open-ended coaxial probe. The Synthetic Minority Oversampling Technique method was adopted to modify the dataset. Feature parameters were scored to select the appropriate feature vector using a Statistical Dependency algorithm. The dataset was classified using adaptive PNNs with an optimized smooth factor using the simulated annealing PNN (SA-PNN). The results were compared with traditional Probabilistic, Support Vector Machines, k-Nearest Neighbor and the Classify functions in MATLAB.ResultsThe conductivity frequencies of 3959, 3958, 3960, 3978, 3510, 3889, 3888, and 3976 MHz were selected as the feature vectors for 219 lymph nodes (178 non-metastatic and 41 metastatic). Compared with the other methods, SA-PNN achieved the highest classification accuracy (92.92%) and the corresponding specificity and sensitivity were 94.72% and 91.11%, respectively.ConclusionsCompared with the other methods, the SA-PNN proposed in the present study achieved a higher classification accuracy, which provides a new scheme for classification of metastatic and non-metastatic thoracic lymph nodes in lung cancer patients based on dielectric properties.