High-quality Features Research Articles

BaGFN: Broad Attentive Graph Fusion Network for High-Order Feature Interactions.

Modeling feature interactions is of crucial significance to high-quality feature engineering on multifiled sparse data. At present, a series of state-of-the-art methods extract cross features in a rather implicit bitwise fashion and lack enough comprehensive and flexible competence of learning sophisticated interactions among different feature fields. In this article, we propose a new broad attentive graph fusion network (BaGFN) to better model high-order feature interactions in a flexible and explicit manner. On the one hand, we design an attentive graph fusion module to strengthen high-order feature representation under graph structure. The graph-based module develops a new bilinear-cross aggregation function to aggregate the graph node information, employs the self-attention mechanism to learn the impact of neighborhood nodes, and updates the high-order representation of features by multihop fusion steps. On the other hand, we further construct a broad attentive cross module to refine high-order feature interactions at a bitwise level. The optimized module designs a new broad attention mechanism to dynamically learn the importance weights of cross features and efficiently conduct the sophisticated high-order feature interactions at the granularity of feature dimensions. The final experimental results demonstrate the effectiveness of our proposed model.

PL-GNet: Pixel Level Global Network for detection and localization of image forgeries

Unlike most Image Forgery Detection and Localization (IFDL) methods that classify the tampered regions by local patch, the features from the whole image in the spatial and frequency domains are leveraged in this paper to classify each pixel in the image. This paper proposes a high-confidence pixel level global network called PL-GNet to combat real-life image forgery that commonly involves different types and visually looks particularly realistic. There are three building blocks in our end-to-end PL-GNet framework: (1) An Encoding net allows us to extract the global features and generate the high-quality feature maps which indicate possible tampered regions. The newly designed first layer and backbone network architecture based on atrous convolution in Encoding net are adopted to capture the changes of pixel relationships and extract rich multi-scale spatial information. (2) A Long Short Term Memory (LSTM) network based on co-occurrence matrix is designed to capture the tampering traces and the discriminative features between manipulated and non-manipulated regions. (3) A Decoding net that incorporates the output of Encoding net and LSTM network learns the mapping from low-resolution feature maps to pixel-wise prediction masks. Furthermore, we conduct a series of ablation experiments to optimize the design of the Encoding network systematically. Extensive experiments on the six challenging datasets demonstrate that our PL-GNet outperforms each subnetwork and consistently achieves state-of-the-art performance compared to alternative methods over three evaluation metrics.

Technological Solutions for Digital Identity

Digital transformation of enterprises is driving the need for a digital identity to recognize people for delivering services. The implementation of digital identity is complex, requiring several technological solutions and much coordination. Capturing and processing data is challenging because biometric issues may arise due to imaging errors. This article addresses this issue and proposes a computer vision-based framework for contactless recognition process using a focus group discussion approach for inputs from experts. The proposed framework enhances image capturing process, extraction of high-quality features from captured images, image processing, contactless face detection, and authentication. The study also derives lessons for other biometric-based identity projects based on image analysis. The proposed framework can be used as a reference for understanding multidimensional perspectives, scalability, and adoption of technological solutions in other similar projects in developing countries in future.

Ultrafast Plane Wave Imaging Using Tensor Completion-Based Minimum Variance Algorithm.

Coherent plane wave compounding (CPWC) imaging is an efficient technique in high-frame-rate ultrasound imaging. To improve the image quality obtained from the CPWC, the adaptive minimum variance (MV) algorithm can be used. However, the high computational complexity of this algorithm negatively affects the frame rate. In other words, achieving a high frame rate and high-quality features simultaneously remains a challenge in medical ultrasound imaging. The aim of the work described here was to develop an algorithm to tackle this challenge and improve the frame rate while preserving the good quality of the resulting image. A tensor completion (TC)-based MV algorithm is proposed to simultaneously improve the frame rate and image quality in CPWC. In the proposed method, the MV algorithm is applied to a limited number of pixels in the beamforming grid. Then, the appropriate values are assigned to the remaining unprocessed pixels by using the TC algorithm. The proposed algorithm speeds up the beamforming process, and consequently, improves the frame rate. The computational complexity of the proposed TC-based MV algorithm is reduced compared with that of the conventional MV algorithm while the good quality of this algorithm is preserved. The results indicate that, in particular, by processing 40% of the beamforming grid using the MV beamformer followed by the TC algorithm, a reconstructed image comparable to that in the case in which the MV algorithm is performed on the full beamforming grid is obtained; the difference between the contrast-to-noise ratio evaluation metric between these two cases is about 0.16 dB for the experimental-resolution phantom. Also, the resulting images obtained from the MV algorithm and the TC-based MV method have the same resolution, indicating that the TC-based MV algorithm can successfully achieve the quality of the MV algorithm with a lower computational complexity. The TC-based MV algorithm is proposed in CPWC with the goal of improving frame rate and image quality. Qualitative and quantitative results reveal that by use of the proposed algorithm, the quality of the reconstructed image will be comparable to that of the conventional MV algorithm, and the frame rate will be improved.

MFDANet: Multi-Scale Feature Dual-Stream Aggregation Network for Salient Object Detection

With the development of deep learning, significant improvements and optimizations have been made in salient object detection. However, many salient object detection methods have limitations, such as insufficient context information extraction, limited interaction modes for different level features, and potential information loss due to a single interaction mode. In order to solve the aforementioned issues, we proposed a dual-stream aggregation network based on multi-scale features, which consists of two main modules, namely a residual context information extraction (RCIE) module and a dense dual-stream aggregation (DDA) module. Firstly, the RCIE module was designed to fully extract context information by connecting features from different receptive fields via residual connections, where convolutional groups composed of asymmetric convolution and dilated convolution are used to extract features from different receptive fields. Secondly, the DDA module aimed to enhance the relationships between different level features by leveraging dense connections to obtain high-quality feature information. Finally, two interaction modes were used for dual-stream aggregation to generate saliency maps. Extensive experiments on 5 benchmark datasets show that the proposed model performs favorably against 15 state-of-the-art methods.

Boosting Semi-Supervised Semantic Segmentation with Probabilistic Representations

Recent breakthroughs in semi-supervised semantic segmentation have been developed through contrastive learning. In prevalent pixel-wise contrastive learning solutions, the model maps pixels to deterministic representations and regularizes them in the latent space. However, there exist inaccurate pseudo-labels which map the ambiguous representations of pixels to the wrong classes due to the limited cognitive ability of the model. In this paper, we define pixel-wise representations from a new perspective of probability theory and propose a Probabilistic Representation Contrastive Learning (PRCL) framework that improves representation quality by taking its probability into consideration. Through modelling the mapping from pixels to representations as the probability via multivariate Gaussian distributions, we can tune the contribution of the ambiguous representations to tolerate the risk of inaccurate pseudo-labels. Furthermore, we define prototypes in the form of distributions, which indicates the confidence of a class, while the point prototype cannot. More- over, we propose to regularize the distribution variance to enhance the reliability of representations. Taking advantage of these benefits, high-quality feature representations can be derived in the latent space, thereby the performance of se- mantic segmentation can be further improved. We conduct sufficient experiment to evaluate PRCL on Pascal VOC and CityScapes to demonstrate its superiority. The code is available at https://github.com/Haoyu-Xie/PRCL.

Minority-Oriented Vicinity Expansion with Attentive Aggregation for Video Long-Tailed Recognition

A dramatic increase in real-world video volume with extremely diverse and emerging topics naturally forms a long-tailed video distribution in terms of their categories, and it spotlights the need for Video Long-Tailed Recognition (VLTR). In this work, we summarize the challenges in VLTR and explore how to overcome them. The challenges are: (1) it is impractical to re-train the whole model for high-quality features, (2) acquiring frame-wise labels requires extensive cost, and (3) long-tailed data triggers biased training. Yet, most existing works for VLTR unavoidably utilize image-level features extracted from pretrained models which are task-irrelevant, and learn by video-level labels. Therefore, to deal with such (1) task-irrelevant features and (2) video-level labels, we introduce two complementary learnable feature aggregators. Learnable layers in each aggregator are to produce task-relevant representations, and each aggregator is to assemble the snippet-wise knowledge into a video representative. Then, we propose Minority-Oriented Vicinity Expansion (MOVE) that explicitly leverages the class frequency into approximating the vicinity distributions to alleviate (3) biased training. By combining these solutions, our approach achieves state-of-the-art results on large-scale VideoLT and synthetically induced Imbalanced-MiniKinetics200. With VideoLT features from ResNet-50, it attains 18% and 58% relative improvements on head and tail classes over the previous state-of-the-art method, respectively. Code and dataset are available at https://github.com/wjun0830/MOVE.

CGC-Net: A Context-Guided Constrained Network for Remote-Sensing Image Super Resolution

In remote-sensing image processing tasks, images with higher resolution always result in better performance on downstream tasks, such as scene classification and object segmentation. However, objects in remote-sensing images often have low resolution and complex textures due to the imaging environment. Therefore, effectively reconstructing high-resolution remote-sensing images remains challenging. To address this concern, we investigate embedding context information and object priors from remote-sensing images into current deep learning super-resolution models. Hence, this paper proposes a novel remote-sensing image super-resolution method called Context-Guided Constrained Network (CGC-Net). In CGC-Net, we first design a simple but effective method to generate inverse distance maps from the remote-sensing image segmentation maps as prior information. Combined with prior information, we propose a Global Context-Constrained Layer (GCCL) to extract high-quality features with global context constraints. Furthermore, we introduce a Guided Local Feature Enhancement Block (GLFE) to enhance the local texture context via a learnable guided filter. Additionally, we design a High-Frequency Consistency Loss (HFC Loss) to ensure gradient consistency between the reconstructed image (HR) and the original high-quality image (HQ). Unlike existing remote-sensing image super-resolution methods, the proposed CGC-Net achieves superior visual results and reports new state-of-the-art (SOTA) performance on three popular remote-sensing image datasets, demonstrating its effectiveness in remote-sensing image super-resolution (RSI-SR) tasks.

A high-quality feature selection method based on frequent and correlated items for text classification

The feature selection problem is a significant challenge in pattern recognition, especially for classification tasks. The quality of the selected features plays a critical role in building effective models, and poor-quality data can make this process more difficult. This work explores the use of association analysis in data mining to select meaningful features, addressing the issue of duplicated information in the selected features. A novel feature selection technique for text classification is proposed, based on frequent and correlated items. This method considers both relevance and feature interactions, using association as a metric to evaluate the relationship between the target and features. The technique was tested using the SMS spam collecting dataset from the UCI machine learning repository and compared with well-known feature selection methods. The results showed that the proposed technique effectively reduced redundant information while achieving high accuracy (95.155%) using only 6% of the features.

Multi-scale attention and boundary enhancement with long-range dependency for salient object detection

Existing saliency detection methods have achieved great progress in extracting multi-level features, however it is a challenging problem to catch accurate long-range dependencies that can enhance the accuracy of semantic information. To address this, a Transformer-based multi-scale attention and boundary enhancement with long-range dependency (MSBE) network is proposed in this paper. A multi-scale attention enhancement module (MSAEM) is designed to reduce the redundant or noisy features and generate a high-quality feature representation by integrating multiple attentional features with diverse perspectives. The high-quality features are then fed into the triple Transformer encoder embedding module (TEM) to enhance high-level semantic features by learning long-range dependencies across layers. In the decoder part, a cross-layer feature fusion module (CLFFM) and boundary enhancement module (BEM) are designed to improve the effect of feature fusion and get accurate prediction results. Extensive experiments on six challenging public datasets demonstrate that the proposed method achieves competitive performance.

Densely Convolutional Spatial Attention Network for nuclei segmentation of histological images for computational pathology.

Automatic nuclear segmentation in digital microscopic tissue images can aid pathologists to extract high-quality features for nuclear morphometrics and other analyses. However, image segmentation is a challenging task in medical image processing and analysis. This study aimed to develop a deep learning-based method for nuclei segmentation of histological images for computational pathology. The original U-Net model sometime has a caveat in exploring significant features. Herein, we present the Densely Convolutional Spatial Attention Network (DCSA-Net) model based on U-Net to perform the segmentation task. Furthermore, the developed model was tested on external multi-tissue dataset - MoNuSeg. To develop deep learning algorithms for well-segmenting nuclei, a large quantity of data are mandatory, which is expensive and less feasible. We collected hematoxylin and eosin-stained image data sets from two hospitals to train the model with a variety of nuclear appearances. Because of the limited number of annotated pathology images, we introduced a small publicly accessible data set of prostate cancer (PCa) with more than 16,000 labeled nuclei. Nevertheless, to construct our proposed model, we developed the DCSA module, an attention mechanism for capturing useful information from raw images. We also used several other artificial intelligence-based segmentation methods and tools to compare their results to our proposed technique. To prioritize the performance of nuclei segmentation, we evaluated the model's outputs based on the Accuracy, Dice coefficient (DC), and Jaccard coefficient (JC) scores. The proposed technique outperformed the other methods and achieved superior nuclei segmentation with accuracy, DC, and JC of 96.4% (95% confidence interval [CI]: 96.2 - 96.6), 81.8 (95% CI: 80.8 - 83.0), and 69.3 (95% CI: 68.2 - 70.0), respectively, on the internal test data set. Our proposed method demonstrates superior performance in segmenting cell nuclei of histological images from internal and external datasets, and outperforms many standard segmentation algorithms used for comparative analysis.

Research on detection method of Tubercle Bacilli based on the improved YOLOv5

In the field of medical imaging, the detection method of Tubercle Bacilli based on deep learning can overcome the shortcomings of traditional manual detection methods, such as large subjectivity, large workload, slow detection speed, and reduce the occurrence of false detection or missed detection under specific circumstances. However, due to the small target and complex background of Tubercle Bacilli, the detection results are still not accurate enough. In order to reduce the influence of sputum sample background on Tubercle Bacilli detection and improve the accuracy of the model for Tubercle Bacilli detection, a target detection algorithm YOLOv5-CTS based on YOLOv5 algorithm is proposed in this paper. The algorithm first integrates the CTR3 module at the bottom of Backbone of YOLOv5 network to obtain more high-quality feature information, which brings significant performance improvement to the model; then in the neck and head part, a hybrid model with the improved feature pyramid networks and the added large-scale detection layer is utilized to perform feature fusion and small target detection; finally, the SCYLLA-Intersection over Union loss function is integrated. The experimental results show that YOLOv5-CTS increases the mean average precision to 86.2% compared with the existing target detection algorithms for Tubercle Bacilli, such as Faster R-CNN, SSD and RetinaNet, etc, which shows the effectiveness of this method.

A Discriminant Information Theoretic Learning Framework for Multi-modal Feature Representation

As sensory and computing technology advances, multi-modal features have been playing a central role in ubiquitously representing patterns and phenomena for effective information analysis and recognition. As a result, multi-modal feature representation is becoming a progressively significant direction of academic research and real applications. Nevertheless, numerous challenges remain ahead, especially in the joint utilization of discriminatory representations and complementary representations from multi-modal features. In this article, a discriminant information theoretic learning (DITL) framework is proposed to address these challenges. By employing this proposed framework, the discrimination and complementation within the given multi-modal features are exploited jointly, resulting in a high-quality feature representation. According to characteristics of the DITL framework, the newly generated feature representation is further optimized, leading to lower computational complexity and improved system performance. To demonstrate the effectiveness and generality of DITL, we conducted experiments on several recognition examples, including both static cases, such as handwritten digit recognition, face recognition, and object recognition, and dynamic cases, such as video-based human emotion recognition and action recognition. The results show that the proposed framework outperforms state-of-the-art algorithms.

Parallel symbolic aggregate approximation and its application in intelligent fault diagnosis

Fault diagnosis is of great significance for industrial equipment maintenance, and feature extraction is a key step of the entire diagnosis scheme. The symbolic aggregate approximation (SAX) is a popular feature extraction approach with great potential recently. In spite of the achievements the SAX has made, the adverse information aliasing still exists in its calculation procedure, and it may make the SAX fail to guarantee the information correctness. This work focuses on analyzing the information aliasing phenomenon of the SAX, followed by developing a novel alternative method, i.e. parallel symbolic aggregate approximation (PSAX). In the proposed PSAX, the information aliasing is suppressed by designing anti-aliasing procedure, and the average of the symbolic results of several intermediate sequence is adopted to replace the final symbolic result. The Case Western Reserve University (CWRU) rolling bearing data together with the gas valve data of an actual reciprocating compressor assist in verifying the superiority exhibited by the suggested method. The experimental results show that, compared with other methods, the accuracy advantage of the PSAX on the 2 datasets can reach 1% –5%, indicating it is capable of providing high-quality feature vector for intelligent fault diagnosis.

Development of Laser Processing Carbon-Fiber-Reinforced Plastic.

Due to its exceptional advantages, such as high specific strength, high specific modulus, and good fatigue resistance, carbon-fiber-reinforced plastic (CFRP) is frequently utilized in aerospace, aviation, automotive, rail transportation, and other areas. Composite components typically need to be joined and integrated. In the equipment manufacturing industry, the most used methods for processing composite components are cutting, drilling, and surface treatment. The quality of CFRP is significantly impacted by traditional mechanical processing, causing flaws like delamination, burrs, and tears. Laser processing technology has emerged as a crucial method for processing CFRP for its high quality, non-contact, simple control, and automation features. The most recent research on the laser processing of CFRP is presented in this paper, supporting scientists and engineers who work in the field in using this unconventional manufacturing technique. This paper gives a general overview of the key features of laser processing technology and the numerous machining techniques available. The concepts and benefits of laser processing technology are discussed in terms of the material properties, mode of operation, and laser characteristics, as well as the methods to achieve high efficiency, low damage, and high precision. This paper reviews the research development of laser processing of carbon-fiber-reinforced plastics, and a summary of the factors affecting the quality of CFRP laser processing. Therefore, the research content of this article can be used as a theoretical basis for reducing thermal damage and improving the processing quality of laser-processed composite materials, while, on this basis, we analyze the development trend of CFRP laser processing technology.

Distributional and spatial-temporal robust representation learning for transportation activity recognition

Transportation activity recognition (TAR) provides valuable support for intelligent transportation applications, such as urban transportation planning, driving behavior analysis, and traffic prediction. There are many advantages of movable sensor-based TAR, and the key challenge is to capture salient features from segmented data for representing diverse patterns of activity. Although existing methods based on statistical information are efficient, they usually rely on domain knowledge to construct high-quality features manually. Likewise, the methods based on spatial-temporal relationships achieve good performance but fail to extract statistical features. The features extracted by these two methods have proven to be crucial for the classification of activity. How to combine them to acquire a more robust representation remains an open question. In this work, we introduce a novel parallel model named Distributional and Spatial-Temporal Robust Representation (DSTRR), which combines automatic learning of statistical, spatial, and temporal features into a unified framework. This model leads to three optimized subnets and thus obtains a robust representation specific to TAR. Extensive experiments performed on three public datasets show that DSTRR is a state-of-the-art method compared with the baseline methods. The results of ablation study and visualization not only demonstrate the effectiveness of each component in DSTRR, but also show the model remains robust to a wide range of parameter variations.

Feature library-assisted surrogate model for evolutionary wrapper-based feature selection and classification

In recent years, wrapper-based feature selection (FS) using evolutionary algorithms has been widely studied due to its ability to search for and evaluate subsets of features based on populations. However, these methods often suffer from a high computational cost and a long computation time, mainly due to the process of evaluating the feature subsets according to the classification performance. In order to tackle this problem, this paper presents a feature library-assisted surrogate model (FL-SM), which aims to reduce the computational cost but maintain a good prediction accuracy. Unlike the existing surrogate models used in FS, the proposed method focuses on the feature level instead of the sample level: an FL is built by collecting the scores of all the features during the evolutionary search. Specifically, each solution (subset candidate) is pre-evaluated based on the FL using only simple operations to decide whether or not it deserves to be evaluated by the classifier, improving the efficiency of the FS algorithm. Meanwhile, because not evaluating a certain number of solutions may lead to inaccurate solution selection during the evolutionary search, dynamic individual selection criteria are proposed. In addition, an adaptive FL update operator is proposed to handle the dynamics of the evolved population; it ensures the real-time validity of the FL. Furthermore, we incorporate the proposed FL-SM into some state-of-the-art single- and multi-objective evolutionary FS methods. The experimental results on benchmark datasets show that with good flexibility and extendibility, FL-SM can effectively reduce the computational cost of wrapper-based FS and still obtain high-quality feature subsets. Among the five algorithms tested, the average computation time reduction was 34.87%; at the same time, there was no significant difference in the classification accuracy for 80% of the tests, and our method even improved the classification accuracy for 6% of the tests.

Dynamic graph convolutional networks by semi-supervised contrastive learning

The traditional graph convolutional network(GCN) and its variants usually only propagate node information through the topology given by the dataset. However, the given topology can only represent a certain relationship and ignore some correlative feature information between nodes, which may make the graph convolutional networks unable to fully utilize the data information. To address the above issue, a novel model named Dynamic Graph Convolutional Networks by Semi-Supervised Contrastive Learning (DGSCL) is proposed in this paper. First, a feature graph is dynamically constructed from the input node features to exploit the potential correlative feature information between nodes. Then, to ensure a high-quality feature graph, a semi-supervised contrastive learning method is designed to learn discriminative node embeddings, which can iteratively refine the constructed feature graph with the learned node embeddings. Finally, we fuse the node embeddings obtained from the given topology and the dynamic feature graph by two co-attention modules to produce more informative embeddings for the classification task. Through a series of experiments, we demonstrate the competitive performance of our model on seven node classification benchmarks.

Preparation of microfluidics device from PMMA for liposome synthesis

Microfluidics has emerged in recent years as a technology that has advantages and is well suited for studying chemistry, biology, and physics at the microscale. A common material which has been widely use to fabricate the microfluidic system is thermoplastic materials. The method of fabricating microfluidic devices has been growing because of advantages such as high-quality feature replication, inexpensiveness, and ease of use. However, the major barrier to the utilization of thermoplastics is the lack of bonding methods for different plastic layers to close the microchannels. Therefore, this study focused on fabricating a microfluidic device on poly(methyl methacrylate) (PMMA) plates by laser engraving. The bonding technique for plastic layers has relied on the application of small amounts of ethanol with conditions of low temperatures (100 ⁰C), and relatively low pressures (5 tons) for 2 minutes. With this technique, the microfluidic device is created to operate stably, without leakage or cracking even under high pressure. The microfluidic device was applied to synthesize liposomes with a 5:1 ratio of syringe pump velocity between water and lipid solution. The size of liposomes after synthesis is 109.64 ± 4.62 nm (mean ± sd) and the PDI is in accordance with standard conditions (PDI < 0.200).

Tree Species Classification Based on Self-Supervised Learning with Multisource Remote Sensing Images

In order to solve the problem of manual labeling in semi-supervised tree species classification, this paper proposes a pixel-level self-supervised learning model named M-SSL (multisource self-supervised learning), which takes the advantage of the information of plenty multisource remote sensing images and self-supervised learning methods. Based on hyperspectral images (HSI) and multispectral images (MSI), the features were extracted by combining generative learning methods with contrastive learning methods. Two kinds of multisource encoders named MAAE (multisource AAE encoder) and MVAE (multisource VAE encoder) were proposed, respectively, which set up pretext tasks to extract multisource features as data augmentation. Then the features were discriminated by the depth-wise cross attention module (DCAM) to enhance effective ones. At last, joint self-supervised methods output the tress species classification map to find the trade-off between providing negative samples and reducing the amount of computation. The M-SSL model can learn more representative features in downstream tasks. By employing the feature cross-fusion process, the low-dimensional information of the data is simultaneously learned in a unified network. Through the validation of three tree species datasets, the classification accuracy reached 78%. The proposed method can obtain high-quality features and is more suitable for label-less tree species classification.