Accurate Pseudo Log Prediction Using Machine Learning Based Automatic Log Regularization and Feature Augmentation Method

Objectives/Scope: The performance of ML-based rock properties prediction from seismic with limited and sparse well data is very often inadequate. To address this limitation, we propose a novel automatic well log regularization (ALR) method with specially designed feature augmentation strategy to improve the prediction accuracy. The effectiveness of ALR method is showcased on field data in Malay basin where we successfully predict elastic logs with 30% higher accuracy, while using only 28% less training dataset. Methods, Procedures, Process: The ALR workflow (Figure 1): (1) feature selection and augmentation; (2) training and prediction and (3) prediction optimizations. The workflow starts with predicting any logs type which are available at training but not in blind wells using standard ML workflow for all blind wells (Step 1-2). Then, these intermediately predicted logs at blind well were jointly used as input features together with seismic-derived attributes using a specially designed feature augmentation strategy (Step 3). Finally, Step 1and 2 are then repeated to predict the elastic logs using these augmented input features. Results, Observations, Conclusions: The ALR method was applied on an oil/gas field data in Malay basin to predict elastic logs (AI and SI) at five blind wells from seismic data only and compared to the standard ML workflow. Two wells were used as training (28% of all data). The prediction performance of standard ML workflow (Figure 2a) is poor and can only capture general mean values of the actual AI/SI logs. The results of ALR workflow (Figure 2b) shows 30% better prediction performance compared to the standard ML workflow. In general, the background and high-resolution trend are well captured, and the overall prediction performance is improved using the new proposed prediction method. There are conceivably two explanations for this result: a) the background (low frequency) trend of the well log is properly reconstructed in ALR using only using seismic data. This could mainly lie in the ability of augmented features in better learning the uncertain reflection-reception relationship between seismic data and elastic logs, as well as the spatial/time-varying property of seismic data; (b) The ability to learn meaningful nonlinear feature relationship between input (feature) and output (label) variables with little or no supervision seems to work properly using specially designed feature augmentation. Novel/Additive Information: The ALR method is an ML-based pseudo log generation from seismic data using specially designed feature augmentation strategy. The novel ALR implementation relaxes the requirement of having a massive amount of high-quality labeled data for training and can therefore be applied in areas with limited well data information. ALR method is proven to be highly accurate for direct elastic logs prediction and can potentially be extended to estimate petrophysical properties from seismic data.

Few-Shot Charge Prediction with Data Augmentation and Feature Augmentation

The task of charge prediction is to predict the charge based on the fact description. Existing methods have a good effect on the prediction of high-frequency charges, but the prediction of low-frequency charges is still a challenge. Moreover, there exist some confusing charges that have relatively similar fact descriptions, which can be easily misjudged. Therefore, we propose a model with data augmentation and feature augmentation for few-shot charge prediction. Specifically, the model takes the text description as the input and uses the Mixup method to generate virtual samples for data augmentation. Then, the charge information heterogeneous graph is introduced, and a novel graph convolutional network is designed to extract distinguishability features for feature augmentation. A feature fusion network is used to effectively integrate the charge graph knowledge into the fact to learn semantic-enhanced fact representation. Finally, the semantic-enhanced fact representation is used to predict the charge. In addition, based on the distribution of each charge, a category prior loss function is designed to increase the contribution of low-frequency charges to the model optimization. The experimental results on real-work datasets prove the effectiveness and robustness of the proposed model.

A Lightweight SE-YOLOv3 Network for Multi-Scale Object Detection in Remote Sensing Imagery

Current state-of-the-art detectors achieved impressive performance in detection accuracy with the use of deep learning. However, most of such detectors cannot detect objects in real time due to heavy computational cost, which limits their wide application. Although some one-stage detectors are designed to accelerate the detection speed, it is still not satisfied for task in high-resolution remote sensing images. To address this problem, a lightweight one-stage approach based on YOLOv3 is proposed in this paper, which is named Squeeze-and-Excitation YOLOv3 (SE-YOLOv3). The proposed algorithm maintains high efficiency and effectiveness simultaneously. With an aim to reduce the number of parameters and increase the ability of feature description, two customized modules, lightweight feature extraction and attention-aware feature augmentation, are embedded by utilizing global information and suppressing redundancy features, respectively. To meet the scale invariance, a spatial pyramid pooling method is used to aggregate local features. The evaluation experiments on two remote sensing image data sets, DOTA and NWPU VHR-10, reveal that the proposed approach achieves more competitive detection effect with less computational consumption.

Self-Recurrent Learning and Gap Sample Feature Synthesis-Based Object Detection Method

Object detection-based deep learning by using the looking and thinking twice mechanism plays an important role in electrical construction work. Nevertheless, the use of this mechanism in object detection produces some problems, such as calculation pressure caused by multilayer convolution and redundant features that confuse the network. In this paper, we propose a self-recurrent learning and gap sample feature fusion-based object detection method to solve the aforementioned problems. The network consists of three modules: self-recurrent learning-based feature fusion (SLFF), residual enhancement architecture-based multichannel (REAML), and gap sample-based features fusion (GSFF). SLFF detects objects in the background through an iterative convolutional network. REAML, which serves as an information filtering module, is used to reduce the interference of redundant features in the background. GSFF adds feature augmentation to the network. Simultaneously, our model can effectively improve the operation and production efficiency of electric power companies’ personnel and guarantee the safety of lives and properties.

Multi-view Feature Augmentation with Adaptive Class Activation Mapping

We propose an end-to-end-trainable feature augmentation module built for image classification that extracts and exploits multi-view local features to boost model performance. Different from using global average pooling (GAP) to extract vectorized features from only the global view, we propose to sample and ensemble diverse multi-view local features to improve model robustness. To sample class-representative local features, we incorporate a simple auxiliary classifier head (comprising only one 1x1 convolutional layer) which efficiently and adaptively attends to class-discriminative local regions of feature maps via our proposed AdaCAM (Adaptive Class Activation Mapping). Extensive experiments demonstrate consistent and noticeable performance gains achieved by our multi-view feature augmentation module.

A Robust Context-Based Deep Learning Approach for Highly Imbalanced Hyperspectral Classification

Hyperspectral imaging is an area of active research with many applications in remote sensing, mineral exploration, and environmental monitoring. Deep learning and, in particular, convolution-based approaches are the current state-of-the-art classification models. However, in the presence of noisy hyperspectral datasets, these deep convolutional neural networks underperform. In this paper, we proposed a feature augmentation approach to increase noise resistance in imbalanced hyperspectral classification. Our method calculates context-based features, and it uses a deep convolutional neuronet (DCN). We tested our proposed approach on the Pavia datasets and compared three models, DCN, PCA + DCN, and our context-based DCN, using the original datasets and the datasets plus noise. Our experimental results show that DCN and PCA + DCN perform well on the original datasets but not on the noisy datasets. Our robust context-based DCN was able to outperform others in the presence of noise and was able to maintain a comparable classification accuracy on clean hyperspectral images.