Search

Tool condition monitoring (TCM) during mechanical cutting is critical for maximising the utilisation of cutting tools and minimising the risk of equipment damage and personnel injury. The demand for highly efficient and sustainable machining in modern industries has led to the development of new processes operating under specific conditions. Real-world datasets obtained under harsh cutting conditions often suffer from intense interference, making the anti-interference capabilities of TCM methods crucial for effective industrial applications. Previous literature reviews on TCM have focused on theoretical methods for monitoring tool wear and breakage. However, reviews of the scientific methodologies and technologies employed in TCM for industrial production are limited. The lack of scientific understanding relevant to the monitoring of cutting tools in industrial production should be addressed urgently. The current data processing, feature dimensionality reduction, and decision-making methods utilised in TCM may not adequately fulfil the real-time and anti-interference demands. The TCM methods also face the challenges of small sample sizes and imbalanced data during real-world dataset processing. Therefore, this study conducts a systematic review of TCM methods to overcome these limitations. First, the theoretical guidelines for the application of TCM methods in industrial production are provided. The sensing system, signal processing, feature dimensionality reduction, and decision-making methods for TCM methods are comprehensively discussed in terms of both their advantages and limitations for applications in industrial production. Considering the effects of real-world datasets with small samples and imbalanced data caused by the harsh environment of a real factory, a systematic presentation is proposed at the data, feature, and decision levels. Finally, the challenges and potential research directions of TCM methods for industrial applications are discussed. A research route for smart factory-oriented machining system management is proposed based on published literature. This review bridges the gap between theoretical research and the industrial application of TCM techniques in industrial production. Prospective research and further development of TCM systems will provide the groundwork for establishing smart factories. [Display omitted] • Principles and techniques of tool condition monitoring are comprehensively reviewed. • Application guidance for tool condition monitoring is provided. • Challenges of small samples and imbalanced data faced by real dataset are addressed. • A research route is proposed for smart-factory-oriented machining system management. [ABSTRACT FROM AUTHOR]

[Display omitted] • A small-data machine learning framework was proposed to aid high-entropy alloy design. • R2 of the approach is 0.854 while maintaining high interpretability and efficiency. • Compressive yield strength, fracture elongation are suggested as key factors. High-entropy alloy (HEA) type energetic structural materials (ESMs) offer exceptional strength, adequate ductility and reactivity upon dynamic loading, thus demonstrating great potentials in pyrotechnic applications. However, the main factors governing their energetic performance remain elusive, primarily attributable to the intricate mechanical-thermal-chemical coupling effects and the inherent challenges of HEA design. To address this, we propose a small-data machine learning framework designed to predict the energetic performance of HEA-type ESMs, employing support vector regression, leave-one-out cross-validation, and principal component analysis (PCA) to effectively manage a small, unevenly distributed, and highly dimensional dataset. Notably, the framework achieved a coefficient of determination (R2) of 0.854 while upholding robust performance, interpretability and computational efficiency. Fracture elongation (ε t) and compressive yield strength (σ cys) were identified as critical features, with σ cys positively influencing performance while both ε t and unit theoretical heat of combustion (UTHC) demonstrated negative effect. Guided by the framework, a series of novel Ti-V-Ta-Zr alloys with the comparable UTHC, velocity (v) and weight (m) but tailored ε t and σ cys were designed and tested. Ti 30 V 30 Ta 30 Zr 10 alloy exhibited a commendable balance of mechanical properties and the smallest mean particle size, aligning with the model predictions and suggesting more thorough energy release during ballistic experiments. [ABSTRACT FROM AUTHOR]

• (1) Deep learning-based approaches for mental load recognition in high-risk works. • (2) Multi-modal Information Fusion and Feature optimization of EEG, PPG, and EOG Signals. • (3) Features extracted by random forest can improve recognition performance. • (4) Deep learning significantly enhances recognition performance in model development. • (5) RF feature selection and 1D-CNN model for superior recognition performance. The psychophysiological state of workers in high-risk industries can affect their work efficiency and overall well-being, representing one of the major causes of catastrophic industrial accidents. Mental load is closely associated with the psychological state of workers. This study explores the potential of deep learning (DL) in classifying and recognizing mental load by collecting physiological signals of workers engaged in hazardous operations. Thirty chemical and safety professionals were selected for data collection, including electroencephalography (EEG), photoplethysmography (PPG), and electrooculography (EOG) signals, and extracting physiological indexes, a combination of four feature selection and extraction techniques is employed for multi-modal information fusion and feature dimensionality reduction. DL and traditional machine learning (ML) techniques are utilized to construct a mental load recognition model, and the performance of the model is evaluated using six reliable metrics. The results show that sensitivity testing successfully induced mental load state, with significant changes in EEG, PPG, and EOG indexes. Feature dimensionality reduction can improve the recognition effect and random forest feature selection (RFS) is the best method. Multi-modal information fusion and feature dimensionality reduction are effective methods for improving mental load recognition. The combination of RFS and 1D-convolutional neural network (1D-CNN) model delivers excellent recognition effectiveness, with an overall accuracy of 90.67% and a high mental load recognition accuracy of 95%. Our research can be applied in engineering production to guide occupational suitability and personnel selection, psychological training, and the adjustment of personnel in critical positions in high-risk industries such as chemical, mining, and construction. [ABSTRACT FROM AUTHOR]

Rice is the main staple food for more than half of the world's population. Consumption of long-stored rice will have adverse effects on the human body. Here, we proposed the near-infrared (NIR) hyperspectral imaging (HSI) technique to distinguish rice from different storage years. Multiplicative Scatter Correction (MSC), Standard Normalize Variate (SNV) and 1st Derivative (1st) were used for the pretreatment of HSI data. In order to reduce dimensional spectral features, Principal Component Analysis (PCA) and t-distributed Stochastic Neighbor Embedding (tSNE) were used for data visualization. Besides, spectral characteristic wavelengths were extracted by competitive adaptive reweighted sampling (CARS) and least absolute shrinkage and selection operator (Lasso). Simultaneously, the textural features of rice were analyzed by Local Binary Pattern (LBP), Gray-Level Co-occurrence Matrix (GLCM) and Tamura algorithms. In order to realize feature fusion of spectra and texture, Support Vector Machine (SVM) model was optimized using the Whale optimization algorithm (WOA) and Extreme Gradient Boosting (XGBoost) model were established based on spectral features and textural features. Compared to other models, feature fusion model showed an excellent result with an accuracy of 98.89%. Experimental results suggested that HSI technology can be served as an effective method for rice detection. • Classification and identification of different storage years of rice using hyperspectral imaging (HSI) technique combined with machine learning. • Principal Component Analysis (PCA) and t-distributed Stochastic Neighbor Embedding (tSNE) algorithms were used to dimensionalize the data after preprocessing. • Extreme Gradient Boosting (XGBoost) model and Support Vector Machine (SVM) model optimized by Whale optimization algorithm (WOA) were used. The results were obtained with better classification accuracy. • Texture data was extracted by Local Binary Pattern (LBP) , Gray-Level Co-occurrence Matrix (GLCM) and Tamura algorithm. [ABSTRACT FROM AUTHOR]

Feature selection is a critical task for optimizing system performance and reducing computational overhead in the context of Internet of Things (IoT) applications. This paper presents a two-stage feature selection approach specifically designed for IoT scenarios. In the first stage, a variety of feature dimensionality reduction techniques are employed to significantly reduce the dimensionality of the original feature set by more than 50%. This process results in the creation of a highly effective feature subset, which serves as a solid foundation for subsequent feature selection in the second stage. In the second stage, feature selection is performed on the feature subset by an evolutionary algorithm to obtain high accuracy. Notably, we propose an improved whale optimization algorithm (WOA-HA), which incorporates several improvement factors such as a chaotic Hénon map mechanism (HMM), adaptive coefficient vector (ACV), and a binary operator. To assess the effectiveness of our approach, we compare the performance of WOA-HA with other evolutionary algorithms in terms of feature selection outcomes. Through extensive experiments, our proposed approach achieves an average accuracy of up to 95.5% on Aalto IoT dataset and 98.8% on RT-IoT 2022 dataset, respectively. Meanwhile, the average number of selected features reduced by about 82.5% on Aalto IoT dataset and about 62.3% in the RT-IoT 2022 dataset, respectively. Our proposed approach consistently outperforms other methods and achieves the best performance on most datasets with higher accuracy and fewer features. • Two-stage feature selection approach is proposed for the high-dimensionality of IoT data. • A novel WOA-HA algorithm is proposed in the second stage. • WOA-HA has a special chaotic initial solution and adaptive strategy. • The proposed approach outperforms other algorithms on Aalto IoT dataset and RT-IoT 2022 dataset. • The first stage of experimentation with the Aalto dataset resulted in a reduction of the feature set dimensionality by over 50%. [ABSTRACT FROM AUTHOR]

High-throughput estimation of soybean yield using unmanned aerial vehicle (UAV) imagery can help improve the efficiency of soybean breeding. Previous studies have mainly focused on the extraction of vegetation indices and texture features from two-dimensional(2D) orthophotos to construct empirical models of yield, lacking spatial structure information of crops. Therefore, UAV cross-circling oblique (CCO) photography combined with SfM-MVS algorithm was used to reconstruct three-dimensional(3D) soybean canopy structure. Then canopy 3D related phenotypic features are extracted and combined with features from RGB nadir and multispectral images to analyze the capability of different modal data fusion on soybean yield prediction. In addition, Shapley value was used to evaluate the importance of features across different machine learning models. Based on the Shapley value, a bagging-stacking ensemble learning framework was developed using Lasso, Random Forest (RF), Ridge Regression (RR), and XGBoost as base learners for yield prediction. The performance of the traditional stacking method was evaluated and compared with weighted average methods such as Bayesian Model Averaging (BMA) and Entropy Weighted Average (EWA) as meta-learners in the ensemble framework. The results show that 3D canopy structure of soybean can be obtained from UAV CCO photography. The inclusion of 3D structural features can improve the accuracy of yield estimation. Among different modal data combinations, the highest estimation accuracy was achieved when combining RGB nadir features with CCO 3D features. The performance of the above base learners was improved by 8.8%, 3.5%, 7.1%, and 8.0% respectively when using the Shapley value method. The accuracy of yield prediction applied on independent dataset of year 2023 was further calculated by using the bagging-stacking ensemble learning method. When using BMA as the meta-learner, the best performance is obtained with an R2 of 0.7. Therefore, UAV CCO photography with SfM-MVS algorithm provides a new approach to obtain high-quality point clouds of the crop canopy at low cost. UAV multimodal data combined with ensemble learning models allow accurate estimation of yield prediction in breeding materials of soybeans at the plot scale. • The potential of predicting soybean yield using multi-source sensors was studied. • UAV Cross-circling oblique photography can be used for soybean three-dimensional canopy reconstruction at the plot scale. • Shapley value has high applicability in feature dimensionality reduction. • Bagging-stacking ensemble learning can be used as a high-precision model in soybean yield modeling. [ABSTRACT FROM AUTHOR]

In this paper, a hybrid system is proposed for speech emotion recognition (SER). The algorithm in this paper adopts a two-stage design concept. In the first stage, we use the ensemble learning model random forest algorithm to obtain the importance of each feature. We use Emo-DB for experimental comparison and find that the combination of the logistic regression algorithm and the WBCS algorithm achieves best results. Cross-training method is used to ensure the features adapt to various situations. Under 100 training sets, the sentiment classification results are satisfactory. The proposed method is more accurate than state-of-the-art intelligent optimization dimensionality reduction algorithms. [ABSTRACT FROM AUTHOR]

• Several major EEG feature extraction methods are introduced. • Several major EEG feature reduction methods are introduced. • Different types of machine learning classifiers for emotion recognition are reviewed. • Several open problems and future research directions in the area of emotion recognition are identified. In recent years, the rapid advances in machine learning (ML) and information fusion has made it possible to endow machines/computers with the ability of emotion understanding, recognition, and analysis. Emotion recognition has attracted increasingly intense interest from researchers from diverse fields. Human emotions can be recognized from facial expressions, speech, behavior (gesture/posture) or physiological signals. However, the first three methods can be ineffective since humans may involuntarily or deliberately conceal their real emotions (so-called social masking). The use of physiological signals can lead to more objective and reliable emotion recognition. Compared with peripheral neurophysiological signals, electroencephalogram (EEG) signals respond to fluctuations of affective states more sensitively and in real time and thus can provide useful features of emotional states. Therefore, various EEG-based emotion recognition techniques have been developed recently. In this paper, the emotion recognition methods based on multi-channel EEG signals as well as multi-modal physiological signals are reviewed. According to the standard pipeline for emotion recognition, we review different feature extraction (e.g., wavelet transform and nonlinear dynamics), feature reduction, and ML classifier design methods (e.g., k-nearest neighbor (KNN), naive Bayesian (NB), support vector machine (SVM) and random forest (RF)). Furthermore, the EEG rhythms that are highly correlated with emotions are analyzed and the correlation between different brain areas and emotions is discussed. Finally, we compare different ML and deep learning algorithms for emotion recognition and suggest several open problems and future research directions in this exciting and fast-growing area of AI. [ABSTRACT FROM AUTHOR]

In this paper we propose a new approach for feature dimensionality reduction based on Reservoir Computing (Echo State Networks). The method is validated with EEG data to identify the common neural signatures based on which the positive and negative valence of human emotions across multiple subjects can be reliably discriminated. The key step in the proposed approach is the Intrinsic Plasticity (IP) adaptation of the reservoir states. Learning Echo State Networks (ESN) with IP maximizes the entropy of the distribution of reservoir vectors given static data as a fixed input, which is supposed to follow Gaussian distribution. The equilibrium reservoir vector is extracted for each static input vector by iterating updates of the reservoir vector until it converges. Standard classification and clustering models provided with selected combinations of reservoir neurons are ranked based on their discriminate performance. The IP tuned ESNs is more powerful technique to map the high dimensional input feature vector into a low dimensional representation and improve the emotion valence discrimination compared to classical ESNs and Deep Neural Encoders. [ABSTRACT FROM AUTHOR]

As robots are progressing towards being ubiquitous and an indispensable part of our everyday environments, such as home, offices, healthcare, education, and manufacturing shop floors, efficient and safe collaboration and cohabitation become imperative. Given that, such environments could benefit greatly from accurate human action prediction. In addition to being accurate, human action prediction should be computationally efficient, in order to ensure a timely reaction, and capable of dealing with changing environments, since unstructured interaction and collaboration with humans usually do not assume static conditions. In this paper, we propose a model for human action prediction based on motion cues and gaze using shared-weight Long Short-Term Memory networks (LSTMs) and feature dimensionality reduction. LSTMs have proven to be a powerful tool in processing time series data, especially when dealing with long-term dependencies; however, to maximize their performance, LSTM networks should be fed with informative and quality inputs. Given that, in this paper, we furthermore conducted an extensive input feature analysis based on (i) signal correlation and their strength to act as stand-alone predictors, and (ii) a multilayer perceptron inspired by the autoencoder architecture. We validated the proposed model on a publicly available MoGaze 1 1 https://humans-to-robots-motion.github.io/mogaze/. dataset for human action prediction, as well as on a smaller dataset recorded in our laboratory. Our model outperformed alternatives, such as recurrent neural networks, a fully connected LSTM network, and the strongest stand-alone signals (baselines), and can run in real-time on a standard laptop CPU. Since eye gaze might not always be available in a real-world scenario, we have implemented and tested a multi-layer perceptron for gaze estimation from more easily obtainable motion cues, such as head orientation and hand position. The estimated gaze signal can be utilized during inference of our LSTM-based model, thus making our action prediction pipeline suitable for real-time practical applications. • Human action prediction using LSTM networks. • Dimensionality reduction using correlation and autoencoder-inspired MLP. • Gaze estimation might improve human action prediction. • Experiments verified approach using motion capture input. [ABSTRACT FROM AUTHOR]

Abstract: Existing feature dimensionality reduction algorithms are inherently designed for the case of classification, clustering and retrieval, but not for ranking applications such as visual search reranking. This is a serious limitation which restricts the applicability of existing dimensionality reduction methods as well as the generalization ability of ranking applications. Therefore, it is important to design a kind of special methods to be effectively employed for ranking applications. Fisher discriminant analysis (FDA) is one of the most popular dimensionality reduction methods. Thus, we propose a novel dimensionality reduction algorithm based on FDA to solve this kind of problem in this paper. Specifically, relevance degree information—the data label in ranking applications, is introduced to a semi-supervised form of FDA, in which both local information and unlabeled data are employed. We name the proposed method as ranking Fisher discriminant analysis (RFDA). To verify the effectiveness of RFDA, extensive experiments are carried out on image search reranking applications, which show significant performance based on the popular MSRA-MM dataset. [Copyright &y& Elsevier]

Abstract: Ranking relevance degree information is widely utilized in the ranking models of information retrieval applications, such as text and multimedia retrieval, question answering, and visual search reranking. However, existing feature dimensionality reduction methods neglect this kind of valuable potential supervised information. In this paper, we extend the pairwise constraints from the traditional class labels to ranking relevance degrees, and propose a novel dimensionality reduction method called Rank-CCA. Rank-CCA effectively incorporates ranking relevance constraints into standard canonical correlation analysis (CCA) algorithm, and is able to employ the knowledge of both unlabeled and labeled data. In the application of visual search reranking, our proposed method is verified through extensive experimental studies. Experimental results show that Rank-CCA is superior to standard CCA and semi-supervised CCA (Semi-CCA) algorithm, and achieves comparable performance with several state-of-the-art reranking methods while preserving the superiority of low dimensional features. [Copyright &y& Elsevier]

• Characterization of manual and non-manual gestures in recognizing the sign gestures. • Deep Networks of self-learning capacity to achieve higher recognition rate. • Iterative optimization on hyperparameters and considered limited training data. • Recognize multimodal and multilingual sign corpus with multi-signer variation. Classifying manual and non-manual gestures in sign language recognition is a complex and challenging task. Sign language gestures are the combination of hand, face, and body postures, which often have self-occlusions and inter-object occlusions of both the hands, hands with face, or hands with upper body postures. This paper addresses the characterization of manual and non-manual gestures in recognizing the sign language gestures from continuous video sequences. This paper introduces a novel hyperparameter based optimized Generative Adversarial Networks (H-GANs) to classify the sign gestures, and it works in three phases. In phase-I, it adapts the stacked variational auto-encoders (SVAE) and Principal Component Analysis (PCA) to get the pre-tuned data with reduced feature dimensions. In Phase-II, the H-GANs employed Deep Long Short Term Memory (LSTM) as generator and LSTM with 3D Convolutional Neural Network (3D-CNN) as a discriminator. The generator generates random sequences with noise from the real sequence of frames, and the discriminator detects and classifies the real frames of sign gestures. In Phase-III, the proposed approach employs Deep Reinforcement Learning (DRL) for hyperparameter optimization and regularization. By getting the reward points, Proximal Policy Optimization (PPO) optimizes the hyperparameters, and Bayesian Optimization (BO) regularizes the hyperparameters. The proposed H-GANs used two different large vocabulary sign corpus of continuous sign videos to evaluate the performance and efficiency of the system. The experimental results on different dimensions reveal that the H-GANs improved the accuracy and recognition rate when compared with the state-of-the-art classification methods with reduced complexity. [ABSTRACT FROM AUTHOR]

• Gastric tumor volume automatically segmented from CT images. • Introducing a new image feature regeneration method in the field of medical imaging informatics to optimize extracted features with higher performance compared to one of the popular feature reduction methods in this field. • Achieving small feature vectors representing the complex and non-linear image feature space. Non-invasively predicting the risk of cancer metastasis before surgery can play an essential role in determining which patients can benefit from neoadjuvant chemotherapy. This study aims to investigate and test the advantages of applying a random projection algorithm to develop and optimize a radiomics-based machine learning model to predict peritoneal metastasis in gastric cancer patients using a small and imbalanced computed tomography (CT) image dataset. A retrospective dataset involving CT images acquired from 159 patients is assembled, including 121 and 38 cases with and without peritoneal metastasis, respectively. A computer-aided detection scheme is first applied to segment primary gastric tumor volumes and initially compute 315 image features. Then, five gradients boosting machine (GBM) models embedded with five feature selection methods (including random projection algorithm, principal component analysis, least absolute shrinkage, and selection operator, maximum relevance and minimum redundancy, and recursive feature elimination) along with a synthetic minority oversampling technique, are built to predict the risk of peritoneal metastasis. All GBM models are trained and tested using a leave-one-case-out cross-validation method. Results show that the GBM model embedded with a random projection algorithm yields a significantly higher prediction accuracy (71.2%) than the other four GBM models (p<0.05). The precision, sensitivity, and specificity of this optimal GBM model are 65.78%, 43.10%, and 87.12%, respectively. This study demonstrates that CT images of the primary gastric tumors contain discriminatory information to predict the risk of peritoneal metastasis, and a random projection algorithm is a promising method to generate optimal feature vector, improving the performance of machine learning based prediction models. [ABSTRACT FROM AUTHOR]