Feature Transformation Research Articles

DNIM: Deep-sea netting intelligent enhancement and exposure monitoring using bio-vision

Intelligent monitoring of deep-sea nets is affected by light attenuation, light scattering, and limited dynamic range factors of the camera, which can cause color shift, low visibility, and over/underexposure in monitoring, resulting in reduced farming efficiency, and misdetection of biological behavior. Therefore, we propose a method for underwater net tank scene enhancement and exposure using multicolor space and bio-vision, called DNIM. Specifically, we first propose an intelligent enhancement and exposure monitoring method for deep-sea nets using bio-vision, which eliminates the effect of exposure on underwater imaging by introducing a bio-vision system that uses a brightness transformation function to generate a series of exposure images. Inspired by the image fusion method, we designed a color correction and visibility recovery method using bio-vision. In the Lab color model, color shifts in bio-vision are corrected by guiding feature transformations with contrast-constrained adaptive histogram equalization; in the RGB color model, depth information is recovered by creating a linear model for the depth of field of underwater blurred images based on the characteristics of the bio-vision system. To obtain more qualified results, we also propose an efficient perceptual fusion module to mix the output of visibility recovery and color correction. Finally, we propose two deep-sea net tank culture datasets: salmon bait-eating behavior dataset (SBBD) and salmon non-bait-eating behavior dataset (SNBD). By comparing the experimental analysis in four real scenario datasets, our DNIM achieved superior results compared to 16 advanced underwater enhancement methods. The code is publicly available at: https://github.com/An-Shunmin/DNIM.

A study of online academic risk prediction based on neural network multivariate time series features

SummaryNeural networks are becoming increasingly widely used in various fields, especially for academic risk forecasts. Academic risk prediction is a hot topic in the field of big data in education that aims to identify and help students who experience great academic difficulties. In recent years, the use of machine learning algorithms and deep learning algorithms to achieve academic risk prediction has garnered increased attention and development. However, most of these studies use nontime series data as features for prediction, which are slightly insufficient in terms of timeliness. Therefore, this article focuses on time series data features that are more expressive of changes in students' learning status and uses multivariate time series data as predictive features. This article proposes a method based on multivariate time series features and a neural network to predict academic risk. The method includes three steps: first, the multivariate time series feature is extracted from the interaction records of the students' online learning platforms; second, the multivariate time series feature transformation model ROCKET is applied to convert the multivariate time series feature into a new feature; third, the new feature is converted into a final prediction result using a linear classification model. Comparative tests show that the proposed method has high effectiveness.

Unveiling the unseen potential of graph learning through MLPs: Effective graph learners using propagation-embracing MLPs

Recent studies attempted to utilize multilayer perceptrons (MLPs) to solve semi-supervised node classification on graphs, by training a student MLP by knowledge distillation (KD) from a teacher graph neural network (GNN). While previous studies have focused mostly on training the student MLP by matching the output probability distributions between the teacher and student models during KD, it has not been systematically studied how to inject the structural information in an explicit and interpretable manner. Inspired by GNNs that separate feature transformation T and propagation Π, we re-frame the KD process as enabling the student MLP to explicitly learn both T and Π. Although this can be achieved by applying the inverse propagation Π−1 before distillation from the teacher GNN, it still comes with a high computational cost from large matrix multiplications during training. To solve this problem, we propose Propagate & Distill (P&D), which propagates the output of the teacher GNN before KD and can be interpreted as an approximate process of the inverse propagation Π−1. Through comprehensive evaluations using real-world benchmark datasets, we demonstrate the effectiveness of P&D by showing further performance boost of the student MLP.

A Linguistically Interpretable Deep Fuzzy Classification System With Feature Transformation and Reconstruction

A Linguistically Interpretable Deep Fuzzy Classification System With Feature Transformation and Reconstruction

Multimodal Retinal Imaging Classification for Parkinson's Disease Using a Convolutional Neural Network.

Changes in retinal structure and microvasculature are connected to parallel changes in the brain. Two recent studies described machine learning algorithms trained on retinal images and quantitative data that identified Alzheimer's dementia and mild cognitive impairment with high accuracy. Prior studies also demonstrated retinal differences in individuals with PD. Herein, we developed a convolutional neural network (CNN) to classify multimodal retinal imaging from either a Parkinson's disease (PD) or control group. We trained a CNN to receive retinal image inputs of optical coherence tomography (OCT) ganglion cell-inner plexiform layer (GC-IPL) thickness color maps, OCT angiography 6 × 6-mm en face macular images of the superficial capillary plexus, and ultra-widefield (UWF) fundus color and autofluorescence photographs to classify the retinal imaging as PD or control. The model consists of a shared pretrained VGG19 feature extractor and image-specific feature transformations which converge to a single output. Model results were assessed using receiver operating characteristic (ROC) curves and bootstrapped 95% confidence intervals for area under the ROC curve (AUC) values. In total, 371 eyes of 249 control subjects and 75 eyes of 52 PD subjects were used for training, validation, and testing. Our best CNN variant achieved an AUC of 0.918. UWF color photographs were the most effective imaging input, and GC-IPL thickness maps were the least contributory. Using retinal images, our pilot CNN was able to identify individuals with PD and serves as a proof of concept to spur the collection of larger imaging datasets needed for clinical-grade algorithms. Developing machine learning models for automated detection of Parkinson's disease from retinal imaging could lead to earlier and more widespread diagnoses.

Cross-domain few-shot semantic segmentation for the astronaut work environment

The study of few-shot semantic segmentation (FSS) for the astronaut work environment (AWE) is of significant importance as it enables the segmentation of unknown categories. However, general FSS methods are predicated on the assumption that the training and testing data belong to the same domain. When this assumption is invalid, the model’s performance is significantly degraded. We propose a more general approach, whereby the model is trained on a generic dataset and tested on a dedicated AWE dataset. This challenging task is referred to as cross-domain few-shot semantic segmentation (CD-FSS). A novel model, namely FTDCNet, is proposed, which comprises a domain-agnostic feature transformation module and a domain-constrained transformer. The FTDCNet model demonstrates superior performance compared to the state-of-the-art (SOTA) model, with an accuracy improvement of 11.83% and 11.42% under 1-shot and 5-shot settings, respectively.

Disentangled Cascaded Graph Convolution Networks for Multi-Behavior Recommendation

Multi-behavioral recommender systems have emerged as a solution to address data sparsity and cold-start issues by incorporating auxiliary behaviors alongside target behaviors. However, existing models struggle to accurately capture varying user preferences across different behaviors and fail to account for diverse item preferences within behaviors. Various user preference factors (such as price or quality) entangled in the behavior may lead to sub-optimization problems. Furthermore, these models overlook the personalized nature of user behavioral preferences by employing uniform transformation networks for all users and items. To tackle these challenges, we propose the Disentangled Cascaded Graph Convolutional Network (Disen-CGCN), a novel multi-behavior recommendation model. Disen-CGCN employs disentangled representation techniques to effectively separate factors within user and item representations, ensuring their independence. In addition, it incorporates a multi-behavioral meta-network, enabling personalized feature transformation across user and item behaviors. Furthermore, an attention mechanism captures user preferences for different item factors within each behavior. By leveraging attention weights, we aggregate user and item embeddings separately for each behavior, computing preference scores that predict overall user preferences for items. Our evaluation of benchmark datasets demonstrates the superiority of Disen-CGCN over state-of-the-art models, showcasing an average performance improvement of 7.07% and 9.00% on respective datasets. These results highlight Disen-CGCN’s ability to effectively leverage multi-behavioral data, leading to more accurate recommendations.

InstLane Dataset and Geometry-Aware Network for Instance Segmentation of Lane Line Detection

Despite impressive progress, obtaining appropriate data for instance-level lane segmentation remains a significant challenge. This limitation hinders the refinement of granular lane-related applications such as lane line crossing surveillance, pavement maintenance, and management. To address this gap, we introduce a benchmark for lane instance segmentation called InstLane. To the best of our knowledge, InstLane constitutes the first publicly accessible instance-level segmentation standard for lane line detection. The complexity of InstLane emanates from the fact that the original data are procured using cameras mounted laterally, as opposed to traditional front-mounted sensors. InstLane encapsulates a range of challenging scenarios, enhancing the generalization and robustness of the lane line instance segmentation algorithms. In addition, we propose GeoLaneNet, a real-time, geometry-aware lane instance segmentation network. Within GeoLaneNet, we design a finer localization of lane proto-instances based on geometric features to counteract the prevalent omission or multiple detections in dense lane scenarios resulting from non-maximum suppression (NMS). Furthermore, we present a scheme that employs a larger receptive field to achieve profound perceptual lane structural learning, thereby improving detection accuracy. We introduce an architecture based on partial feature transformation to expedite the detection process. Comprehensive experiments on InstLane demonstrate that GeoLaneNet can achieve up to twice the speed of current State-Of-The-Artmethods, reaching 139 FPS on an RTX3090 and a mask AP of 73.55%, with a permissible trade-off in AP, while maintaining comparable accuracy. These results underscore the effectiveness, robustness, and efficiency of GeoLaneNet in autonomous driving.

A Method for Reducing Training Time of ML-Based Cascade Scheme for Large-Volume Data Analysis.

We live in the era of large data analysis, where processing vast datasets has become essential for uncovering valuable insights across various domains of our lives. Machine learning (ML) algorithms offer powerful tools for processing and analyzing this abundance of information. However, the considerable time and computational resources needed for training ML models pose significant challenges, especially within cascade schemes, due to the iterative nature of training algorithms, the complexity of feature extraction and transformation processes, and the large sizes of the datasets involved. This paper proposes a modification to the existing ML-based cascade scheme for analyzing large biomedical datasets by incorporating principal component analysis (PCA) at each level of the cascade. We selected the number of principal components to replace the initial inputs so that it ensured 95% variance retention. Furthermore, we enhanced the training and application algorithms and demonstrated the effectiveness of the modified cascade scheme through comparative analysis, which showcased a significant reduction in training time while improving the generalization properties of the method and the accuracy of the large data analysis. The improved enhanced generalization properties of the scheme stemmed from the reduction in nonsignificant independent attributes in the dataset, which further enhanced its performance in intelligent large data analysis.

ANN-assisted prediction of wave run-up around a tension leg platform under irregular wave conditions

This study proposes a novel prediction method for nonlinear wave run-ups around a TLP by combining the linear diffraction method with deep learning techniques. Initially, the linear diffraction method evaluates relative wave motion around the platform based on measured incident waves and frequency domain analysis. Subsequently, an artificial neural network (ANN) predicts the peak values of nonlinear wave run-ups, utilizing incident wave data and linear calculation results as inputs. Two different ANN models were proposed, each tailored to specific output variables. The first, the ζ-based ANN model, predicts the peak value of wave run-ups, while the second, the α-based ANN model, predicts the nonlinear amplification factor. The incorporation of a neural network significantly enhances the prediction accuracy for the peak values of wave run-ups. Moreover, the α-based ANN model demonstrates superior accuracy in predicting high run-up events and better generalization ability across various wave conditions compared to the ζ-based ANN model. The study also investigates the effect of various preprocessing methods on prediction performance. The over-sampling method improves accuracy for high run-up events but degrades predictions for low run-up regions, while the polynomial feature transformation reduces prediction errors for high run-up in the ζ-based ANN model.

Securing Machine Learning: Understanding Adversarial Attacks and Bias Mitigation

This paper offers a comprehensive examination of adversarial vulnerabilities in machine learning (ML) models and strategies for mitigating fairness and bias issues. It analyses various adversarial attack vectors encompassing evasion, poisoning, model inversion, exploratory probes, and model stealing, elucidating their potential to compromise model integrity and induce misclassification or information leakage. In response, a range of defence mechanisms including adversarial training, certified defences, feature transformations, and ensemble methods are scrutinized, assessing their effectiveness and limitations in fortifying ML models against adversarial threats. Furthermore, the study explores the nuanced landscape of fairness and bias in ML, addressing societal biases, stereotypes reinforcement, and unfair treatment, proposing mitigation strategies like fairness metrics, bias auditing, de-biasing techniques, and human-in-the-loop approaches to foster fairness, transparency, and ethical AI deployment. This synthesis advocates for interdisciplinary collaboration to build resilient, fair, and trustworthy AI systems amidst the evolving technological paradigm.

Semantic guidance incremental network for efficiency video super-resolution

AbstractIn video streaming, bandwidth constraints significantly affect client-side video quality. Addressing this, deep neural networks offer a promising avenue for implementing video super-resolution (VSR) at the user end, leveraging advancements in modern hardware, including mobile devices. The principal challenge in VSR is the computational intensity involved in processing temporal/spatial video data. Conventional methods, uniformly processing entire scenes, often result in inefficient resource allocation. This is evident in the over-processing of simpler regions and insufficient attention to complex regions, leading to edge artifacts in merged regions. Our innovative approach employs semantic segmentation and spatial frequency-based categorization to divide each video frame into regions of varying complexity: simple, medium, and complex. These are then processed through an efficient incremental model, optimizing computational resources. A key innovation is the sparse temporal/spatial feature transformation layer, which mitigates edge artifacts and ensures seamless integration of regional features, enhancing the naturalness of the super-resolution outcome. Experimental results demonstrate that our method significantly boosts VSR efficiency while maintaining effectiveness. This marks a notable advancement in streaming video technology, optimizing video quality with reduced computational demands. This approach, featuring semantic segmentation, spatial frequency analysis, and an incremental network structure, represents a substantial improvement over traditional VSR methodologies, addressing the core challenges of efficiency and quality in high-resolution video streaming.

Identification and Removal of Pollen Spectral Interference in the Classification of Hazardous Substances Based on Excitation Emission Matrix Fluorescence Spectroscopy.

Sensitively detecting hazardous and suspected bioaerosols is crucial for safeguarding public health. The potential impact of pollen on identifying bacterial species through fluorescence spectra should not be overlooked. Before the analysis, the spectrum underwent preprocessing steps, including normalization, multivariate scattering correction, and Savitzky-Golay smoothing. Additionally, the spectrum was transformed using difference, standard normal variable, and fast Fourier transform techniques. A random forest algorithm was employed for the classification and identification of 31 different types of samples. The fast Fourier transform improved the classification accuracy of the sample excitation-emission matrix fluorescence spectrum data by 9.2%, resulting in an accuracy of 89.24%. The harmful substances, including Staphylococcus aureus, ricin, beta-bungarotoxin, and Staphylococcal enterotoxin B, were clearly distinguished. The spectral data transformation and classification algorithm effectively eliminated the interference of pollen on other components. Furthermore, a classification and recognition model based on spectral feature transformation was established, demonstrating excellent application potential in detecting hazardous substances and protecting public health. This study provided a solid foundation for the application of rapid detection methods for harmful bioaerosols.

RCFNet: Related Cross-level Feature Network with Cascaded Self-distillation for Monocular Depth Estimation

Monocular depth estimation (MDE) is a challenging yet crucial computer vision task, which aims to generate accurate depth maps from a single image. Existing MDE approaches mainly rely on extracting and fusing diverse information from multi-level features to improve prediction accuracy. However, these methods often apply a traditional feature pyramid structure, neglecting a comprehensive exploration of feature fusion paths across multiple levels. Moreover, a single-feature fusion strategy has limited ability to optimize the network. We propose a novel related cross-level feature network (RCFNet) with cascaded self-distillation for monocular depth estimation, including a cross-level feature enhancement (CLFE) module, a hierarchical feature cross refinement (HFCR) module, and a cascaded self-distillation (CSD) module. The CLFE module integrates cross-level features to further exploit the highest-level features, where a channel attention mechanism with hybrid weight operations is deployed to enhance the initial features. The HFCR module adaptively captures strongly correlated complementary information through a window-based multi-head cross-attention mechanism to generate refined features. Meanwhile, the CSD module with a hierarchical feature transformation loss is proposed, which can be viewed as a virtual teacher to progressively extract discriminative features within the network for better gradient flow improvement. Extensive experiments on NYUv2 and KITTI datasets demonstrate that our method outperforms existing SOTA MDE methods in terms of accuracy capacity and robustness.

A knowledge-data integration framework for rolling element bearing RUL prediction across its life cycle

Prediction of Remaining Useful Life (RUL) for Rolling Element Bearings (REB) has attracted widespread attention from academia and industry. However, there are still several bottlenecks, including the effective utilization of multi-sensor data, the interpretability of prediction models, and the prediction across the entire life cycle, which limit prediction accuracy. In view of that, we propose a knowledge-based explainable life-cycle RUL prediction framework. First, considering the feature fusion of fast-changing signals, the Pearson correlation coefficient matrix and feature transformation objective function are incorporated to an Improved Graph Convolutional Autoencoder. Furthermore, to integrate the multi-source signals, a Cascaded Multi-head Self-attention Autoencoder with Characteristic Guidance is proposed to construct health indicators. Then, the whole life cycle of REB is divided into different stages based on the Continuous Gradient Recognition with Outlier Detection. With the development of Measurement-based Correction Life Formula and Bidirectional Recursive Gated Dual Attention Unit, accurate life-cycle RUL prediction is achieved. Data from self-designed test rig and PHM 2012 Prognostic challenge datasets are analyzed with the proposed framework and five existing prediction models. Compared with the strongest prediction model among the five, the proposed framework demonstrates significant improvements. For the data from self-designed test rig, there is a 1.66 % enhancement in Corrected Cumulative Relative Accuracy (CCRA) and a 49.00 % improvement in Coefficient of Determination (R2). For the PHM 2012 datasets, there is a 4.04 % increase in CCRA and a 120.72 % boost in R2.

Correction to: Multi-View Feature Transformation Based SVM+ for Computer-Aided Diagnosis of Liver Cancers With Ultrasound Image.

Presents corrections to the paper, Multi-View Feature Transformation Based SVM+ for Computer-Aided Diagnosis of Liver Cancers With Ultrasound Image.

SPD Manifold Deep Metric Learning for Image Set Classification.

By characterizing each image set as a nonsingular covariance matrix on the symmetric positive definite (SPD) manifold, the approaches of visual content classification with image sets have made impressive progress. However, the key challenge of unhelpfully large intraclass variability and interclass similarity of representations remains open to date. Although, several recent studies have mitigated the two problems by jointly learning the embedding mapping and the similarity metric on the original SPD manifold, their inherent shallow and linear feature transformation mechanism are not powerful enough to capture useful geometric features, especially in complex scenarios. To this end, this article explores a novel approach, termed SPD manifold deep metric learning (SMDML), for image set classification. Specifically, SMDML first selects a prevailing SPD manifold neural network (SPDNet) as the backbone (encoder) to derive an SPD matrix nonlinear representation. To counteract the degradation of structural information during multistage feature embedding, we construct a Riemannian decoder at the end of the encoder, trained by a reconstruction error term (RT), to induce the generated low-dimensional feature manifold of the hidden layer to capture the pivotal information about the visual data describing the imaged scene. We demonstrate through theory and experiments that it is feasible to replace the Riemannian metric with Euclidean distance in RT. Then, the ReCov layer is introduced into the established Riemannian network to regularize the local statistical information within each input feature matrix, which enhances the effectiveness of the learning process. The theoretical analysis of the activation function used in the ReCov layer in terms of continuity and conditions for generating positive definite matrices is beneficial for network design. Inspired by the fact that the single cross-entropy loss used for training is unable to effectively parse the geometric distribution of the deep representations, we finally endow the suggested model with a novel metric learning regularization term. By explicitly incorporating the encoding and processing of the data variations into the network learning process, this term can not only derive a powerful Riemannian representation but also train an effective classifier. The experimental results show the superiority of the proposed approach on three typical visual classification tasks.

X-Detect: explainable adversarial patch detection for object detectors in retail

Object detection models, which are widely used in various domains (such as retail), have been shown to be vulnerable to adversarial attacks. Existing methods for detecting adversarial attacks on object detectors have had difficulty detecting new real-life attacks. We present X-Detect, a novel adversarial patch detector that can: (1) detect adversarial samples in real time, allowing the defender to take preventive action; (2) provide explanations for the alerts raised to support the defender’s decision-making process, and (3) handle unfamiliar threats in the form of new attacks. Given a new scene, X-Detect uses an ensemble of explainable-by-design detectors that utilize object extraction, scene manipulation, and feature transformation techniques to determine whether an alert needs to be raised. X-Detect was evaluated in both the physical and digital space using five different attack scenarios (including adaptive attacks) and the benchmark COCO dataset and our new Superstore dataset. The physical evaluation was performed using a smart shopping cart setup in real-world settings and included 17 adversarial patch attacks recorded in 1700 adversarial videos. The results showed that X-Detect outperforms the state-of-the-art methods in distinguishing between benign and adversarial scenes for all attack scenarios while maintaining a 0% FPR (no false alarms) and providing actionable explanations for the alerts raised. A demo is available.

A novel dimensionality reduction method based on flow model

The sparsity and redundancy of high-dimensional data poses challenges during processing. Dimensionality reduction is a commonly used approach, which helps to reduce data redundancy and noise, lower the complexity of learning algorithms, and improve classification accuracy. Typically, directly determining the intrinsic dimensionality of a dataset is challenging. Many of the present dimensionality reduction techniques suffer from pre-set target dimensions without considering the specific dataset features. To address this problem, a novel dimensionality reduction method based on flow model, which is called adaptive flow encoder (AFE), is proposed in this paper. Initially, the flow model is employed to achieve feature transformation and project the data onto an equip-dimensional feature space. Subsequently, a gating layer is utilized for feature selection. By configuring the gating layer parameters to be learnable, the model gains the ability to adaptively select features. To assess the efficacy of the proposed approach, the model’s ability to preserve data characteristics is first qualitatively evaluated through visual inspection by projecting the data onto a two-dimensional space. Subsequently, five standard datasets are employed and compared against six traditional dimensionality reduction techniques to ascertain the impact on downstream classification tasks under the adaptive learning of intrinsic feature dimensions by the model. Moreover, the impact of key parameters, including the initialization scheme of the gate layer, amplification factor, and dimension factor, on the test results is examined.

On the effectiveness of hybrid pooling in mixup-based graph learning for language processing

Graph neural network (GNN)-based graph learning has been popular in natural language and programming language processing, particularly in text and source code classification. Typically, GNNs are constructed by incorporating alternating layers which learn transformations of graph node features, along with graph pooling layers that use graph pooling operators (e.g., Max-pooling) to effectively reduce the number of nodes while preserving the semantic information of the graph. Recently, to enhance GNNs in graph learning tasks, Manifold-Mixup, a data augmentation technique that produces synthetic graph data by linearly mixing a pair of graph data and their labels, has been widely adopted. However, the performance of Manifold-Mixup can be highly affected by graph pooling operators, and there have not been many studies that are dedicated to uncovering such affection. To bridge this gap, we take an early step to explore how graph pooling operators affect the performance of Mixup-based graph learning. To that end, we conduct a comprehensive empirical study by applying Manifold-Mixup to a formal characterization of graph pooling based on 11 graph pooling operations (9 hybrid pooling operators, 2 non-hybrid pooling operators). The experimental results on both natural language datasets (Gossipcop, Politifact) and programming language datasets (JAVA250, Python800) demonstrate that hybrid pooling operators are more effective for Manifold-Mixup than the standard Max-pooling and the state-of-the-art graph multiset transformer (GMT) pooling, in terms of producing more accurate and robust GNN models.Editor’s note: Open Science material was validated by the Journal of Systems and Software Open Science Board.