Feature Pool Research Articles

A multi-step fast charging-based battery capacity estimation framework of real-world electric vehicles

Accurately evaluating battery degradation is not only crucial for ensuring the safe and reliable operation of electric vehicles (EVs) but also fundamental for their intelligent management and maximum utilization. However, the non-linearity, non-measurability, and multi-stress coupled operating conditions have posed significant challenges for battery health prediction. This paper proposes a battery capacity estimation framework based on real-world operating data. Firstly, a comprehensive feature pool is constructed from the direct external features extracted during multi-step fast charging processes and the quantitative representation of operating conditions. Subsequently, a two-step feature engineering is introduced to select the most relevant features and eliminate the interference components. The battery capacity estimation framework is then implemented using machine learning methods. Validation results demonstrate that the proposed framework achieves superior estimation accuracy with lower computational expense compared to the modelling process without feature engineering. The MAPE and RMSE reach 1.18% and 1.98 Ah, respectively, representing reductions in errors of up to 8.53% and 11.21%. Collectively, the proposed framework paves the foundation for online health prognostics of batteries under practical operating conditions.

Compressive mechanical response and microstructures in low strain rate plastic deformation of stainless steel 316L fabricated by selective laser melting

Characterization of the mechanical properties plays an essential role in the post-processing and evaluation of the functionality of the additively manufactured metallic parts. A number of studies have been focused on the tensile properties of additively manufactured metals. However, the quasi-static compression test of the additively manufactured 316L blocks with different heat-treatment conditions and scanning strategies seems to be overlooked in the literature. This paper aims to provide a comprehensive study of compressive mechanical response in plastic deformation of SS316L fabricated by selective laser melting (SLM). The mechanical response and microstructures in compressive deformation is analyzed for three printing strategies with 0°-, 90°- and 67.5°- scanning and three heat treatment conditions (450 °C for 3 h, 1100 °C for 1 h with furnace cool and 1100 °C for 1 h with water quenching) for selective laser melted (SLMed) stainless steel 316L in comparison with wrought stainless steel 316L in this work. Alteration of mechanical properties, microstructure evolution and compressive deformation mechanism is studied. Melt pool features are not significantly affected by low-temperature heat treatment (450 °C for 3 h) but fully dissolved through high-temperature heat treatment (1100 °C for 1 h). High-temperature heat treatment provides a higher resistance to compressive plastic deformation for SLMed 316L compared with the low-temperature heat-treated and as-built samples where more twinnings are observed. The compressive plastic deformation mechanism of 90°- and 67.5°-scanning samples is similar, which mainly results from twinning-induced plasticity. For 0°-scanning samples, the strong crystallographic texture is the main cause of anisotropic deformation. Modelling and simulation have been conducted to explain the anisotropic deformation mechanism of the 0°-scanning strategy. Simulation results suggest that the morphology difference of laser-scanning tracks and melt pools, which leads to material flow along the laser scanning direction, explains the anisotropic deformation mechanism.

Classification of Typical Static Objects in Road Scenes Based on LO-Net

Mobile LiDAR technology is a powerful tool that accurately captures spatial information about typical static objects in road scenes. However, the precise extraction and classification of these objects pose persistent technical challenges. In this paper, we employ a deep learning approach to tackle the point cloud classification problem. Despite the popularity of the PointNet++ network for direct point cloud processing, it encounters issues related to insufficient feature learning and low accuracy. To address these limitations, we introduce a novel layer-wise optimization network, LO-Net. Initially, LO-Net utilizes the set abstraction module from PointNet++ to extract initial local features. It further enhances these features through the edge convolution capabilities of GraphConv and optimizes them using the “Unite_module” for semantic enhancement. Finally, it employs a point cloud spatial pyramid joint pooling module, developed by the authors, for the multiscale pooling of final low-level local features. Combining three layers of local features, LO-Net sends them to the fully connected layer for accurate point cloud classification. Considering real-world scenarios, road scene data often consist of incomplete point cloud data due to factors such as occlusion. In contrast, models in public datasets are typically more complete but may not accurately reflect real-world conditions. To bridge this gap, we transformed road point cloud data collected by mobile LiDAR into a dataset suitable for network training. This dataset encompasses nine common road scene features; hence, we named it the Road9 dataset and conducted classification research based on this dataset. The experimental analysis demonstrates that the proposed algorithm model yielded favorable results on the public datasets ModelNet40, ModelNet10, and the Sydney Urban Objects Dataset, achieving accuracies of 91.2%, 94.2%, and 79.5%, respectively. On the custom road scene dataset, Road9, the algorithm model proposed in this paper demonstrated outstanding classification performance, achieving a classification accuracy of 98.5%.

A Novel Approach to Impact Crater Mapping and Analysis on Enceladus, Using Machine Learning

AbstractImpact cratering is one of the most important processes shaping planetary surfaces, offering valuable clues about the target body's geologic history and composition. However, crater mapping has historically been done manually, a process that has proven to be both arduous and time consuming. This paper outlines a machine learning crater mapping approach for bodies with limited elevation data available (Digital Elevation Models). We applied a Convolutional Neural Network for the detection and morphometry of impact craters on Saturn's moon Enceladus using light‐shadow labels trained on data from the Cassini Imaging Science Subsystem. Our algorithm identified a total of 5,240 features which were used to quantify crater distribution; this included the highest number of small craters (<1–2 km in diameter) recorded on Enceladus by any previous published study. The pool of features was later down‐selected to craters between 0 and 30°N (latitude) imaged at high incidence (>60°) and phase angles (>26°). The down selection was necessary to accurately perform diameter measurements and derive depths from shadow estimation techniques to calculate depth–diameter ratios (d/D); a well‐studied relationship used to constrain planetary surface properties. Results show that the d/D ratio of craters in the equatorial region of Enceladus range from ∼0.06 to 0.37, with a median of 0.19. Our results will inform efforts to constrain the surface properties of this region of Enceladus, potentially also supporting future mission concept design for the Saturnian moon. Future work will explore the simple‐to‐complex crater transition and differences between this area's d/D and Enceladus' northern and southern latitudes.

Effect of supplementary private health insurance on out-of-pocket inpatient medical expenditure: evidence from Malaysia.

The vital role of healthcare financing in achieving universal health coverage is indisputable. However, most countries, including Malaysia, face challenges in establishing an equitable and sustainable healthcare financing system due to escalating healthcare costs, an ageing population and a growing disease burden. With desirable pre-payment and risk pooling features, private health insurance (PHI) is considered an alternative financing option to reduce out-of-pocket (OOP) medical expenditure. However, ongoing theoretical and empirical debates persist regarding the adequacy of financial risk protection provided by PHI largely because it depends on its role, the benefit design and the regulations in place. Our study aimed to investigate the effect of supplementary PHI on OOP inpatient medical expenditure in Malaysia. Secondary data analysis was conducted using the Malaysian National Health and Morbidity Survey 2019 dataset. A total of 983 respondents with a history of inpatient hospitalization in the past 12 months were included in the study. Instrumental variable analysis using a two-stage residual inclusion was performed to address endogeneity bias, with wealth status and education level as the instrumental variables. Tobit regression model was used in the second stage considering the censored distribution of the outcome variable. Missing data were handled using multiple imputation. About one-fifth of the respondents had PHI. In this study, we found that having PHI significantly increased OOP inpatient medical expenditure in all three marginal effects. Additionally, age, residential location, ethnicity (citizenship), being covered by government guarantee letter, government funding and employer-sponsored health insurance were other significant factors associated with OOP inpatient medical expenditure. Our findings undermine a key justification to advocate PHI uptake among the population, with a need for the Malaysian government to reassess the role of PHI in healthcare financing and reconsider PHI subsidization policy. Regulations should also be strengthened to enhance the financial risk protection provided by PHI.

Fault Diagnosis in Centrifugal Pumps: A Dual-Scalogram Approach with Convolution Autoencoder and Artificial Neural Network

This paper proposes a new fault diagnosis method for centrifugal pumps by combining signal processing with deep learning techniques. Centrifugal pumps facilitate fluid transport through the energy generated by the impeller. Throughout the operation, variations in the fluid pressure at the pump’s inlet may impact the generalization of traditional machine learning models trained on raw statistical features. To address this concern, first, vibration signals are collected from centrifugal pumps, followed by the application of a lowpass filter to isolate frequencies indicative of faults. These signals are then subjected to a continuous wavelet transform and Stockwell transform, generating two distinct time–frequency scalograms. The Sobel filter is employed to further highlight essential features within these scalograms. For feature extraction, this approach employs two parallel convolutional autoencoders, each tailored for a specific scalogram type. Subsequently, extracted features are merged into a unified feature pool, which forms the basis for training a two-layer artificial neural network, with the aim of achieving accurate fault classification. The proposed method is validated using three distinct datasets obtained from the centrifugal pump under varying inlet fluid pressures. The results demonstrate classification accuracies of 100%, 99.2%, and 98.8% for each dataset, surpassing the accuracies achieved by the reference comparison methods.

Dealing with the Data Imbalance Problem in Pulsar Candidate Sifting Based on Feature Selection

Pulsar detection has become an active research topic in radio astronomy recently. One of the essential procedures for pulsar detection is pulsar candidate sifting (PCS), a procedure for identifying potential pulsar signals in a survey. However, pulsar candidates are always class-imbalanced, as most candidates are non-pulsars such as RFI and only a tiny part of them are from real pulsars. Class imbalance can greatly affect the performance of machine learning (ML) models, resulting in a heavy cost as some real pulsars are misjudged. To deal with the problem, techniques of choosing relevant features to discriminate pulsars from non-pulsars are focused on, which is known as feature selection. Feature selection is a process of selecting a subset of the most relevant features from a feature pool. The distinguishing features between pulsars and non-pulsars can significantly improve the performance of the classifier even if the data are highly imbalanced. In this work, an algorithm for feature selection called the K-fold Relief-Greedy (KFRG) algorithm is designed. KFRG is a two-stage algorithm. In the first stage, it filters out some irrelevant features according to their K-fold Relief scores, while in the second stage, it removes the redundant features and selects the most relevant features by a forward greedy search strategy. Experiments on the data set of the High Time Resolution Universe survey verified that ML models based on KFRG are capable of PCS, correctly separating pulsars from non-pulsars even if the candidates are highly class-imbalanced.

A Tracking-Based Two-Stage Framework for Spatio-Temporal Action Detection

Spatio-temporal action detection (STAD) is a task receiving widespread attention and has numerous application scenarios, such as video surveillance and smart education. Current studies follow a localization-based two-stage detection paradigm, which exploits a person detector for action localization and a feature processing model with a classifier for action classification. However, many issues occur due to the imbalance between task settings and model complexity in STAD. Firstly, the model complexity of heavy offline person detectors adds to the inference overhead. Secondly, the frame-level actor proposals are incompatible with the video-level feature aggregation and Region-of-Interest feature pooling in action classification, which limits the detection performance under diverse action motions and results in low detection accuracy. In this paper, we propose a tracking-based two-stage spatio-temporal action detection framework called TrAD. The key idea of TrAD is to build video-level consistency and reduce model complexity in our STAD framework by generating action track proposals among multiple video frames instead of actor proposals in a single frame. In particular, we utilize tailored tracking to simulate the behavior of human cognitive actions and used the captured motion trajectories as video-level proposals. We then integrate a proposal scaling method and a feature aggregation module into action classification to enhance feature pooling for detected tracks. Evaluations in the AVA dataset demonstrate that TrAD achieves SOTA performance with 29.7 mAP, while also facilitating a 58% reduction in overall computation compared to SlowFast.

Weighted Combination of Łukasiewicz implication and Fuzzy Jaccard similarity in Hybrid Ensemble Framework (WCLFJHEF) for Gene Selection

A framework is developed for gene expression analysis by introducing fuzzy Jaccard similarity (FJS) and combining Łukasiewicz implication with it through weights in hybrid ensemble framework (WCLFJHEF) for gene selection in cancer. The method is called weighted combination of Łukasiewicz implication and fuzzy Jaccard similarity in hybrid ensemble framework (WCLFJHEF). While the fuzziness in Jaccard similarity is incorporated by using the existing Gödel fuzzy logic, the weights are obtained by maximizing the average F-score of selected genes in classifying the cancer patients. The patients are first divided into different clusters, based on the number of patient groups, using average linkage agglomerative clustering and a new score, called WCLFJ (weighted combination of Łukasiewicz implication and fuzzy Jaccard similarity). The genes are then selected from each cluster separately using filter based Relief-F and wrapper based SVMRFE (Support Vector Machine with Recursive Feature Elimination). A gene (feature) pool is created by considering the union of selected features for all the clusters. A set of informative genes is selected from the pool using sequential backward floating search (SBFS) algorithm. Patients are then classified using Naïve Bayes’(NB) and Support Vector Machine (SVM) separately, using the selected genes and the related F-scores are calculated. The weights in WCLFJ are then updated iteratively to maximize the average F-score obtained from the results of the classifier. The effectiveness of WCLFJHEF is demonstrated on six gene expression datasets. The average values of accuracy, F-score, recall, precision and MCC over all the datasets, are 95%, 94%, 94%, 94%, and 90%, respectively. The explainability of the selected genes is shown using SHapley Additive exPlanations (SHAP) values and this information is further used to rank them. The relevance of the selected gene set are biologically validated using the KEGG Pathway, Gene Ontology (GO), and existing literatures. It is seen that the genes that are selected by WCLFJHEF are candidates for genomic alterations in the various cancer types. The source code of WCLFJHEF is available at http://www.isical.ac.in/~shubhra/WCLFJHEF.html.

A highly accurate prediction for heart failure disease: a new deep attentive model with guided feature ranking

Heart failure (HF) is a life-threatening disease affecting at least 64 million people worldwide. Hence, it places great stresses on patients and healthcare systems. Accordingly, providing a computerized model for HF prediction will help in enhancing diagnosis, treatment, and long-term management of HF. In this paper, we introduce a new guided attentive HF prediction approach. In this method, a sparse-guided feature ranking method is proposed. Firstly, a Gauss–Seidel strategy is applied to the preprocessed feature pool for low-rank approximation procedure with a trace-norm regularization. The resultant sparse attributes, after a Spearman ranking elimination, are employed to guide the original feature pool through linear translation-variant model. Then, a fast Newton-based method is employed for a non-negative matrix factorization for the guided feature pool. The resultant bases of the factorization process are finally utilized in the adopted deep attentive predictive model. For the final prediction stage, instead of the commonly used machine learning approaches, we introduce an attentive-based classifier. It employs sequential attention to choose the most proper salient features for efficient interpretability and learning process. For the evaluation of the proposed HF prediction model, three different datasets are employed, i.e., UCI, Faisalabad, and Framingham datasets. Compared to state-of-the-art techniques, the proposed approach outperforms their performance on all datasets with even small feature sizes. With only four feature bases, the proposed method achieves an average accuracy of 98%, while, with full feature bases, full accuracy is gained.

Printability disparities in heterogeneous material combinations via laser directed energy deposition: a comparative study

Additive manufacturing provides achievability for the fabrication of bimetallic and multi-material structures; however, the material compatibility and bondability directly affect the parts’ formability and final quality. It is essential to understand the underlying printability of different material combinations based on an adapted process. Here, the printability disparities of two common and attractive material combinations (nickel- and iron-based alloys) are evaluated at the macro and micro levels via laser directed energy deposition (DED). The deposition processes were captured using in situ high-speed imaging, and the dissimilarities in melt pool features and track morphology were quantitatively investigated within specific process windows. Moreover, the microstructure diversity of the tracks and blocks processed with varied material pairs was comparatively elaborated and, complemented with the informative multi-physics modeling, the presented non-uniformity in mechanical properties (microhardness) among the heterogeneous material pairs was rationalized. The differences in melt flow induced by the unlike thermophysical properties of the material pairs and the resulting element intermixing and localized re-alloying during solidification dominate the presented dissimilarity in printability among the material combinations. This work provides an in-depth understanding of the phenomenological differences in the deposition of dissimilar materials and aims to guide more reliable DED forming of bimetallic parts.

Pedestrian Intrusion Detection in Railway Station Based on Mirror Translation Attention and Feature Pooling Enhancement

Pedestrian Intrusion Detection in Railway Station Based on Mirror Translation Attention and Feature Pooling Enhancement

3D TUMOR SEGMENTATION AND DIAGNOSIS SYSTEM OF LUNG CANCER USING MULTISCALE APPROACHES

A deadly disease that affects people in various countries in the world is lung cancer (LC). The rate at which people die due to LC is high because it cannot be detected easily at its initial stage of tumor development. The lives of many people who are affected by LC are assured if it is detected in the initial stage. The diagnosis of LC is possible with conventional computer-aided diagnosis (CAD). The process of diagnosis can be improved by providing the associated evaluation outcomes to the radiologists. Since the results from the process of extraction of features and segmentation of lung nodule are crucial in determining the operation of the traditional CAD system, the results from the CAD system highly depend on these processes. The LC classification from computed tomography (CT) images of three dimensions (3D) using a CAD system is the key aspect of this paper. The collection of the 3D-CT images from the standard data source takes place in the first stage. The obtained images are provided as input for the segmentation stage, in which a multi-scale 3D TransUNet (M-3D-TUNet) is adopted to get the precise segmentation of the LC images. A multi-cascaded model that incorporates residual network (ResNet), visual geometry group (VGG)-19, and DenseNet models is utilized to obtain the deep features from the segmented images. The segmented image from the M-3D-TUNet model is given as input to this multi-cascaded network. The features are obtained and fused to form the feature pool. The feature-pool features are provided to the enhanced long short-term memory with attention mechanism (ELSTM-AM) for classification of the LC. The ELSTM-AM classifies the images as normal or healthy segments. The classifier's parameters are optimized with the help of the modified fluctuation-based queuing search algorithm (MF-QSA). The output from implementing the suggested model on 3D-CT images from Lung Nodule Analysis of 2016, with a sample of 888 CT scans with 1186 nodules dataset, achieved; Accuracy 90.9%, Precision 91.1%, Sensitivity 91%, Specificity 90.8%, and F-Score 91%, which shows that the generated framework for LC detection is better than existing models for LC classification.

Low-voltage AC series arc fault detection based on Fisher-mutual information feature selection

The detection of multi-feature fusion is a crucial approach to address the issue of series arc fault detection. Effective feature selection plays a vital role in enhancing the accuracy of the classifier and reducing system complexity. In this study, a feature selection algorithm based on Fisher-mutual information is proposed to tackle the problem of feature selection in multi-feature fusion detection. This algorithm utilizes the characteristics of arc fault voltage source to construct a feature pool. The Fisher-score algorithm and mutual information algorithm are employed to construct an optimal feature subset. The feature subset undergoes rough selection by retaining key features of the classifier and fine selection by eliminating redundant features. Experimental results and comparisons with related methods demonstrate that the proposed feature selection method significantly enhances the classifier's recognition accuracy, reduces classification and recognition time, diminishes the feature dimension, and outperforms other existing methods.

LRDNet: Lightweight LiDAR Aided Cascaded Feature Pools for Free Road Space Detection

Humans have long fantasized about self-driving vehicles for the sake of luxury, style, safety, and ease. Free road space detection for collision avoidance and path planning is a vital part of autonomous driving vehicles. Despite many researchers focusing on free road space detection, it remains an open and challenging problem for real-world applications. Many studies have attempted to fuse depth and LiDAR features with visual features to improve the overall performance of free road space detection. However, there is no guideline on how such features should be fused to complement the visual features. Additionally, most of the previously proposed methods are computationally expensive and not suitable for real-life applications. The main motivation of this study is to realize a lightweight model that addresses these problems without compromising performance. As the LiDAR and visual features exist in different spaces, the proposed method attempts to learn various transformation and fusion operations from LiDAR features to complement the visual features. To validate the performance of the proposed method, we conduct comprehensive experiments on prominent benchmark datasets. The results of the experiments reveal the superior performance of the proposed model while being lightweight. LRDNet ranks third overall (with a minor difference) and second among LiDAR-based methods on the KITTI road benchmark dataset. Furthermore, the proposed model is the least computationally expensive among state-of-the-art methods and can be considered an optimal trade-off between speed and accuracy.

An Improved VGG-19 Network Induced Enhanced Feature Pooling for Precise Moving Object Detection in Complex Video Scenes

An Improved VGG-19 Network Induced Enhanced Feature Pooling for Precise Moving Object Detection in Complex Video Scenes

Over-deposition assessment of Direct Energy Deposition (DED) using melt pool geometric features and Machine Learning

Over-deposition’s influence on the DED process is significant in terms of process stability and final part quality. Most recent studies on quality assessment are based on online monitoring of the melt pool and comparing it with pre-specified thresholds. In this study an online machine learning model is developed, capable of identifying two over-deposition defect types, satellite and wire-oscillation, by using frames captured by an optical camera targeting the melt pool. The features extracted from the images included five geometrical melt pool features and the operating laser power and were successfully used to achieve a 94.4% classification accuracy regardless of the final part geometry.

KPDet: Keypoint-based 3D object detection with Parametric Radius Learning

3D object detection requires accurate recognition of object category, size, rotation angle, and location in 3D space. Currently, many 3D object detection methods rely on the compact Bird’s Eye View (BEV) or point-wise representations to generate proposals. However, these proposal generation paradigms neglect the spatial distribution of objects, which causes difficulty in estimating the centers of objects. In this paper, we propose a keypoint-based 3D object detector, KPDet, which employs the keypoints in the neighborhood of object centers as the representation for proposal generation. The proposed KPDet first uses voxel-keypoint mapping to aggregate informative features on the subsampled keypoints from voxel-wise features, then calibrates the misalignment between the keypoints and object centers through the Object-aware Feature Pooling (OFP) module. These aligned keypoints with their corresponding features are applied to generate proposals. Since the keypoints are essential components, we further present the Structural Point Abstraction (SPA) module, which captures the anisotropic features of keypoints via constructed structural points to enhance the geometric information. In addition, based on the well-studied multi-task learning framework, we also propose a Parametric Radius Learning (PRL) strategy to adjust the sampling radius of the point-based feature aggregation process during the training procedure. Extensive experiments on the KITI and Waymo Open Dataset show that KPDet could achieve promising results compared with previous works.

Fractal pooling: A new strategy for texture recognition using convolutional neural networks

Texture recognition is an important task in computer vision and, as most problems in the area nowadays, has benefited from the use of deep convolutional neural networks. Nevertheless, typical architectures designed for object recognition usually do not perform optimally in such tasks. One of the most important elements of deep architectures in this application concerns the pooling operation. In particular, the well established global average pooling fails to capture more complex, multiscale and non-linear relation among the output activations. In this context, we propose fractal pooling, where the average is replaced by fractal dimension of the feature map. The new module is coupled with a convolutional backbone and a trainable residual block. The method is evaluated on texture classification, both on benchmark databases and on a real-world problem in botany. Our results are competitive with the state-of-the-art in texture classification, outperforming several modern deep learning approaches in terms of classification accuracy with the addition of minimal computational burden. The results suggest the potential of the proposed methodology for texture recognition in general and that using complexity measures is a promising strategy to perform pooling of deep features for this type of image.

Pancreatic cancer pathology image segmentation with channel and spatial long-range dependencies

Based on deep learning, pancreatic cancer pathology image segmentation technology effectively assists pathologists in achieving improved treatment outcomes. However, compared to traditional image segmentation tasks, the large size of tissues in pathology images requires a larger receptive field. While methods based on dilated convolutions or attention mechanisms can enhance the receptive field, they cannot capture long-range feature dependencies. Directly applying self-attention mechanisms to capture long-range dependencies results in intolerable computational complexity. To address these challenges, we introduce a channel and spatial self-attention (CS) Module designed for efficiently capturing both channel and spatial long-range feature dependencies in pancreatic cancer pathological images. Specifically, the channel and spatial self-attention module consists of an adaptive channel self-attention module and a window-shift spatial self-attention module. The adaptive channel self-attention module adaptively pools features to a fixed size to capture long-range feature dependencies. While the window-shift spatial self-attention module captures spatial long-range dependencies in a window-based manner. Additionally, we propose a re-weighted cross-entropy loss to mitigate the impact of long-tail distribution on performance. Our proposed method surpasses state-of-the-art on both our Pancreatic Cancer Pathology Image (PCPI) dataset and the GlaS challenge dataset. The mDice and mIoU have achieved 73.93% and 59.42% in our PCPI dataset.