Showing papers on "Feature (computer vision) published in 2022"
TL;DR: A new approach for fault detection and diagnosis in rotating machinery is proposed, namely: unsupervised classification and root cause analysis, and a comparison between models used in machine learning explainability: SHAP and Local Depth-based Feature Importance for the Isolation Forest (Local-DIFFI).
94 citations
TL;DR: A feedback convolutional neural network architecture, including three repeated segments of convolution, pooling, and flattening, is designed in this article for multiple extraction of spectral features from one-dimensional NIR data to ensure high-confidence NIR analysis in the artificial intelligence performance of IoT.
Abstract: Near-infrared (NIR) data containing spectral response information for detecting target composition are sparsely implied in spectral frequency sequence. Spectral feature information should be extracted using computer-oriented chemometric methods. An Internet of Things (IoT) framework constructed with NIR calibration platform needs some advanced algorithm architectures to realize intelligent analysis. A feedback convolutional neural network (CNN) architecture, including three repeated segments of convolution, pooling, and flattening, is designed in this article for multiple extraction of spectral features from one-dimensional NIR data. An error-feedback iteration mechanism is proposed in the model training process to optimize convolution filters of each segment. Multisegment features are fused successively to ease the sparse information issue. Fusion data are further used to train the calibration models with a parametric-scaling fully connected network to determine the suitable numbers of hidden and output nodes. The adaptive network structure has the advantage of obtaining optimal prediction results from fused feature data. The proposed feedback CNN architecture based on feature information fusion is applied to the NIR rapid quantitative detection of selenium content in paddy rice samples. Experimental results showed that the fusion of multisegment features can enhance the ability of spectral information extraction. The optimal model based on fused feature data performs better than models based on separate feature data of each segment. The feedback convolutional network for information fusion can be applied in the NIR collaborative IoT framework for rapid detection spectroscopy to ensure high-confidence NIR analysis in the artificial intelligence performance of IoT.
50 citations
TL;DR: This article proposes a more reasonable and feasible method based on a dual-view scene with calibration-free gaze direction that is feasible and better than the state-of-the-art methods based on multiple widely used metrics.
Abstract: Driver attention estimation is one of the key technologies for intelligent vehicles The existing related methods only focus on the scene image or the driver's gaze or head pose The purpose of this article is to propose a more reasonable and feasible method based on a dual-view scene with calibration-free gaze direction According to human visual mechanisms, the low-level features, static visual saliency map, and dynamic optical flow information are extracted as input feature maps, which combine the high-level semantic descriptions and a gaze probability map transformed from the gaze direction A multiresolution neural network is proposed to handle the calibration-free features The proposed method is verified on a virtual reality experimental platform that collected more than 550 000 samples and obtained a more accurate ground truth The experiments show that the proposed method is feasible and better than the state-of-the-art methods based on multiple widely used metrics This study also provides a discussion of the effects of different landscapes, times, and weather conditions on the performance
48 citations
TL;DR: In this article, a polarization fusion network with geometric feature embedding (PFGFE-Net) was proposed to solve the two defects of traditional feature abandonment and insufficient utilization of traditional features.
46 citations
TL;DR: Wang et al. as mentioned in this paper proposed a multi-task ViT that leverages low-level CXR feature corpus obtained from a backbone network that extracts common CXRs findings.
38 citations
TL;DR: Experimental results indicate that the proposed feature selection algorithm for label distribution learning was more effective than five state-of-art feature selection algorithms on twelve datasets, with respect to six representative evaluation measures.
35 citations
TL;DR: In this article, the authors propose to add visual attention maps as new data alongside images, thus introducing human visual attention into the DNNs training and compare it with both global and local automatic attention mechanisms.
33 citations
TL;DR: In this paper, a novel framework for feature selection that relies on boosting, or sample re-weighting, to select sets of informative features in classification problems is proposed, which uses as its basis the feature rankings derived from fast and scalable tree-boosting models, such as XGBoost.
Abstract: As dimensions of datasets in predictive modelling continue to grow, feature selection becomes increasingly practical. Datasets with complex feature interactions and high levels of redundancy still present a challenge to existing feature selection methods. We propose a novel framework for feature selection that relies on boosting, or sample re-weighting, to select sets of informative features in classification problems. The method uses as its basis the feature rankings derived from fast and scalable tree-boosting models, such as XGBoost. We compare the proposed method to standard feature selection algorithms on 9 benchmark datasets. We show that the proposed approach reaches higher accuracies with fewer features on most of the tested datasets, and that the selected features have lower redundancy.
33 citations
TL;DR: In this article, each feature is normalized independently with one of the methods from the pool of normalization methods, which is in contrast to the conventional approach which normalizes the data with one method only and as a result, yields suboptimal performance.
32 citations
TL;DR: A novel automated classification algorithm by fusing a number of deep learning approaches has been proposed to detect prostate cancer from ultrasound (US) and MRI images and explains why a specific decision is made given the input US or MRI image.
32 citations
TL;DR: Although the lifespan of LIBs is estimated using the training set in the 5 % SOH range, the estimation errors of the proposed framework are less than 2.5 % in all test sets, ensuring its potential applicability in practical implementations of onboard battery management systems.
TL;DR: The LSTM model trained with the combination of surface-riser-downhole comprehensive detection technologies performs the best in reducing both the prediction error and detection time delay, which could be used to quantitatively evaluate the downhole gas kick risk in the more accurate, faster, more stable, more reliable, and cost-effective manner.
TL;DR: An artificial intelligence-assisted distributed system for manufacturing plant-wide predictive maintenance applications that relies on the feature selection technique to identity an optimal feature subset for each type of fault and is enabled by deploying each independent model built on the obtained feature subset into different edge nodes.
Abstract: The emergence of Industry 4.0 and the rapid advances in the Industrial Internet of Things (IIoT) have provided manufacturers with the ability to remotely monitor the process by deploying automatic fault detection in an IoT-based predictive maintenance system. However, the monitoring targets are now manufacturing plant-wide instead of being just a local area. Multiple types of faults are involved and the conventional centralized cloud computing-based IoT solutions always lead to a heavy burden on the network bandwidth due to the large amount of sensor data collected frequently that has to be transmitted to the central server and this leads to poor response time for the monitoring system. To address this problem, this article develops an artificial intelligence-assisted distributed system for manufacturing plant-wide predictive maintenance applications. The developed distributed system relies on the feature selection technique to identity an optimal feature subset for each type of fault and is enabled by deploying each independent model built on the obtained feature subset into different edge nodes. The distributed approach enables the data to be processed near the sensors, requiring less data to be transmitted to the central cloud server reducing network delay and delivering more accurate results. In addition, our proposed feature selection approach is especially designed to accommodate the characteristics of IIoT data such as the lack of labels. The effectiveness of the proposed method is validated using the widely used public Tennessee Eastman dataset.
TL;DR: Zhang et al. as mentioned in this paper proposed to learn a hybrid ranking representation for person re-id with a two-stream architecture: (1) in the external stream, they use the ranking list of each probe image to learn plausible visual variations among the top ranks from the gallery as the external ranking information; (2) In the internal stream, the part-based fine-grained feature as the internal ranking information, which mitigates the harm of incorrect matches in the ranking lists.
TL;DR: In this article, the authors investigated the efficiency of using 17 commonly pre-trained CNN architectures as feature extractors and of 24 machine learning classifiers to evaluate the classification of skin lesions from two different datasets: ISIC 2019 and PH2.
Abstract: For various forms of skin lesion, many different feature extraction methods have been investigated so far. Indeed, feature extraction is a crucial step in machine learning processes. In general, we can distinct handcrafted and deep learning features. In this paper, we investigate the efficiency of using 17 commonly pre-trained convolutional neural networks (CNN) architectures as feature extractors and of 24 machine learning classifiers to evaluate the classification of skin lesions from two different datasets: ISIC 2019 and PH2. In this research, we find out that a DenseNet201 combined with Fine KNN or Cubic SVM achieved the best results in accuracy (92.34% and 91.71%) for the ISIC 2019 dataset. The results also show that the suggested method outperforms others approaches with an accuracy of 99% on the PH2 dataset.
TL;DR: A hybrid deep transfer learning-based approach to PD classification could lead to hitting rates higher than 99%.
TL;DR: In this article, the authors proposed to transform the features extracted by a pre-trained self-supervised feature extractor into a Gaussian-like distribution to reduce the feature distribution mis-match.
TL;DR: A multi-attention augmented network, which mainly consists of content-, orientation- and position-aware modules, is proposed, which develops an attention augmented U-net structure to form the content-aware module in order to learn and combine multi-scale informative features within a large receptive field.
TL;DR: In this article, two methods to optimize the modeling parameters of multi-point geostatistics were proposed based on gray level co-occurrence matrix and convolutional neural network.
TL;DR: An integrated deep multiscale feature fusion network (IDMFFN) for aeroengine RUL prediction using multisensor data is proposed in this article, where a GRU-based high-level feature fusion block is built to replace the traditional fully connected layer and can leverage powerful temporal feature learning for feature fusion.
Abstract: Most RUL prediction methods can only extract single-scale features, ignoring significant details at other scales and layers. These methods are all constructed using one type of model, and do not use the advantages of different models. An integrated deep multiscale feature fusion network (IDMFFN) for aeroengine RUL prediction using multisensor data is proposed in this study. Two-dimensional samples are constructed using multisensor data with multiple time cycles. Multiscale feature extraction blocks are designed to learn different-scale features using convolutional filters of different sizes. A multiscale feature concatenated block is constructed to integrate multiscale features from different layers. A GRU-based high-level feature fusion block is built to replace the traditional fully connected layer, and can leverage powerful temporal feature learning for feature fusion. A novel activation function Mish is used to construct the network. A simulated turbofan engine dataset was used to verify the effectiveness of the network. The results suggest that the IDMFFN can predict RUL more accurately than existing methods.
TL;DR: Wang et al. as discussed by the authors proposed a sewing defect detection method using a CNN feature map extracted from the initial layers of a pre-trained VGG-16 to detect a broken stitch from a captured image of a sewing operation.
Abstract: The inspection of sewing defects is an essential step in the quality assurance of garment manufacturing. Although traditional automated defect detection applications have shown good performance, these methods are usually configured with handcrafted features designed by a human operator. Recently, deep learning methods that include Convolutional Neural Networks (CNNs) have demonstrated excellent performance in a wide variety of computer-vision applications. To take advantage of the CNN’s feature representation, the direct utilization of feature maps from the convolutional layers as universal feature descriptors has been studied. In this paper, we propose a sewing defect detection method using a CNN feature map extracted from the initial layers of a pre-trained VGG-16 to detect a broken stitch from a captured image of a sewing operation. To assess the effectiveness of the proposed method, experiments were conducted on a set of sewing images, including normal images, their synthetic defects, and rotated images. As a result, the proposed method detected true defects with 92.3% accuracy. Moreover, additional conditions for computing devices and deep learning libraries were investigated to reduce the computing time required for real-time computation. Using a general and cheap single-board computer with resizing the image and utilizing a lightweight deep learning library, the computing time was 0.22 s. The results confirm the feasibility of the proposed method’s performance as an appropriate manufacturing technology for garment production.
TL;DR: A simple but effective two-stage spatial pooling process: rich descriptor extraction and information fusion is introduced, which aims to obtain a set of diverse deep descriptors that contain more informative cues than global average-pooling.
TL;DR: Extensive results obtained on the IEEE 39-bus system as well as a large-scale power system in China with field PMU measurements show that the proposed method achieves the highest classification accuracy and computational efficiency as compared to other deep learning algorithms.
Abstract: Data quality issues exist in practical phasor measurement units (PMUs) due to communication errors or signal interferences. As a result, the performances of existing data-driven disturbance classification methods can be significantly affected. In this article, a fast disturbance classification method that is robust to PMU data quality issues is proposed. The impacts of bad PMU measurements on disturbance classification are investigated by analyzing the feature distributions of deep learning methods. A new feature extraction scheme that uses the univariate temporal convolutional denoising autoencoder (UTCN-DAE) is proposed. It allows encoding and decoding univariate disturbance data through a temporal convolutional network to capture the temporal feature representation and is robust to bad data. Based on the features of the frequency and voltage measurements encoded by the UTCN-DAE, a two-stream enhanced network, i.e., the multivariable temporal convolutional denoising network is proposed to achieve optimal feature extraction of multivariate time series by feature fusion. The classification is performed using a multilayered deep neural network and Softmax classifier. Extensive results obtained on the IEEE 39-bus system as well as a large-scale power system in China with field PMU measurements show that the proposed method achieves the highest classification accuracy and computational efficiency as compared to other deep learning algorithms.
TL;DR: Wang et al. as mentioned in this paper proposed a 3D sphere representation-based center-points matching detection network (SCPM-Net) that is anchor-free and automatically predicts the position, radius, and offset of nodules without manual design of nodule/anchor parameters.
TL;DR: In this paper, a brain tumor classification method using the fusion of deep and shallow features is proposed to distinguish between meningioma, glioma and pituitary tumor types and to predict the 1p/19q co-deletion status of LGG tumors.
TL;DR: A novel UFS approach is proposed by integrating local linear embedding (LLE) and manifold regularization constrained in feature subspace into a unified framework, and a tailored iterative algorithm based on Alternative Direction Method of Multipliers (ADMM) is designed to solve the proposed optimization problem.
TL;DR: Wang et al. as mentioned in this paper proposed a class feature attention mechanism fused with an improved Deeplabv3+ network called CFAMNet for semantic segmentation of common features in remote sensing images.
TL;DR: In this paper, a region-aware latent features fusion based clustering (RLFFC) method was proposed to reduce the band redundancy of hyperspectral images (HSIs) by employing superpixel segmentation to segment HSIs into multiple regions so that the spatial information of HSIs can be fully preserved.
TL;DR: Wang et al. as discussed by the authors proposed a heuristic algorithm based on permutation importance (PIMP) to modify the biased feature importance measure, which not only improves accuracy but also makes the results more interpretable.
Abstract: Corporate financial distress prediction research has been ongoing for more than half a century, during which many models have emerged, among which ensemble learning algorithms are the most accurate. Most of the state-of-the-art methods of recent years are based on gradient boosted decision trees. However, most of them do not consider using feature importance for feature selection, and a few of them use the feature importance method with bias, which may not reflect the true importance of features. To solve this problem, a heuristic algorithm based on permutation importance (PIMP) is proposed to modify the biased feature importance measure in this paper. This method ranks and filters the features used by machine learning models, which not only improves accuracy but also makes the results more interpretable. Based on financial data from 4,167 listed companies in China between 2001 and 2019, the experiment shows that compared with using the random forest (RF) wrapper method alone, the bias in feature importance is indeed corrected by combining the PIMP method. After the redundant features are removed, the performance of most machine learning models is improved. The PIMP method is a promising addition to the existing financial distress prediction methods. Moreover, compared with traditional statistical learning models and other machine learning models, the proposed PIMP-XGBoost offers higher prediction accuracy and clearer interpretation, making it suitable for commercial use.
TL;DR: Experimental results on a large-scale 3D indoor point cloud dataset S3DIS and a part-segmentation dataset ShapeNet have demonstrated the superiority of the proposed JSPNet method over existing state-of-the-arts in both semantic and instance segmentation.