Electrical activity is generated in the forearm muscles during muscular contractions that control dexterous movements of a human finger and thumb. Using this electrical activity as an input to train a neural network for the purposes of classifying finger movements is not straightforward. Low cost wearable sensors i.e., a Myo Gesture control armband (www.bynorth.com), generally have a lower sampling rate when compared with medical grade EMG detection systems e.g., 200 Hz vs 2000 Hz. Using sensors such as the Myo coupled with the lower amplitude generated by individual finger movements makes it difficult to achieve high classification accuracy. Low sampling rate makes it challenging to distinguish between large quantities of subtle finger movements when using a single network. This research uses two networks which enables for the reduction in the number of movements in each network that are being classified; in turn improving the classification. This is achieved by developing and training LSTM networks that focus on the extension and flexion signals of the fingers and a separate network that is trained using thumb movement signal data. By following this method, this research have increased classification of the individual finger movements to between 90 and 100%.

This paper presents the development of a 6-dof parallel robot for the performance of assembly tasks in a human-robot collaborative environment. The architecture and design of the robot are selected such that the robot is mechanically backdrivable. Thereby, the robot can physically interact with an environment or with humans without requiring the use of a force/torque sensor, which is the main objective of this work. The architecture of the robot is first described and its kinematic model is established. The Jacobian matrices are derived and an algorithm is presented for the determination of its workspace. The force capabilities of the robot are then established based on a proposed formulation. A prototype of the robot is presented and control schemes are developed, including a controller based on a vision system. Finally, a video demonstrating the experimental validation of the robot accompanies this paper. The video qualitatively demonstrates the interaction capabilities of the robot.

The human lung is a complex organ with high cellular heterogeneity, and its development and maintenance require interactive gene networks and dynamic cross-talk among multiple cell types. We focus on the confocal immunofluorescent (IF) images of lung tissues from the LungMAP database to reveal lung development. Using the current state-of-the-art deep learning-based model, the authors consider obtaining accurate multi-class segmentation of lung confocal IF images. One of the primary bottlenecks in using deep Convolutional Neural Network (CNN) models is the lack of availability of large-scale training or ground-truth segmentation labels. Then, we implement the multi-class segmentation with Generative Adversarial Network (GAN) models to expand the training dataset, improve overall segmentation accuracy, and discuss the effectiveness of created synthetic images in the segmentation of IF images. Consequently, experimental results indicated that 15.1% increased the accuracy of six-class segmentation using Mask R-CNN. In particular, the accuracy of our few data was mainly improved by using our proposed method. Therefore, the synthetic dataset can moderate the imbalanced data and be used for expanding the dataset.

Under a new wave of technological revolution and industrial change, Internet plus has penetrated into every aspect of production and life. For the traditional takeout industry, the new crowdsourcing distribution mode based on o2o (online to offline) provides new ideas for distribution but also leads to great challenges. In view of the existing development problems and transformation needs of the delivery network, this paper explores the delivery mechanism of delivery orders in crowdsourcing mode, focusing on the delivery path and work efficiency. To optimize the delivery network, taking the shortest delivery path and the least time delay as objectives, this paper establishes a crowdsourcing delivery path optimization model with a time window and includes an example application model. The results show that the model can solve the delivery problem in crowdsourcing mode.

Privacy concerns at the individual and public or private organizational levels are a crucial. Its importance is highly evident nowadays, with the development of advanced technology. This study proposes a system for text mining that analyzes characteristics related to language. This factor makes it possible to generate a fictitious system while analyzing the patent within a bird’s-eye view and presenting keywords to support an idea. By mapping each patent’s information and relationship to an n-dimensional space, one can search for similar patents employing cosine similarity. Quantitative and qualitative evaluation verified the usefulness of the system.

Genetic folding (GF) is a robust evolutionary optimization algorithm. For efficient hyper-scale GFs, a hybrid parallel approach based on CPU architecture Parallel GF (PGF) is proposed. It aids in resolving kernel tricks that are difficult to predict using conventional optimization approaches. The regression and classification problems are solved using PGF. Four concurrent CPUs are formed to parallelize the GF, and each executes eight threads. It is also easily scalable to multi-core CPUs. PGFLibPy is a Python-based machine learning framework for classification and regression problems. PGFLibPy was used to build a model of the UCI dataset that reliably predicts regression values. The toolbox activity is used for binary and multiclassification datasets to classify UCI. PGFLibPy’s has 25 Python files and 18 datasets. Dask parallel implementation is being considered in the toolbox. According to this study, this toolbox can categorize and predict models on any other dataset. The source code, binaries, and dataset are available for download at https://github.com/mohabedalgani/PGFLibPy.

This paper focuses on the analysis of the activity of immune cells for supporting medical workers. Recognition frequency space selects a region including neighboring multiple cells as a single cell is one of the major issues in activity analysis of immune cells. This study focuses on the locality of immune cell features and uses a high-velocity weighting method for the analysis while the Gaussian distribution is used in the literature. The analysis was conducted for a few well-known methods such as final feature maps, class activation mapping (CAM), gradient weighted class activation mapping (Grad-CAM), Grad-CAM++, and Eigen-CAM. The results show that the densely inhabited immune cells are correctly selected by CAM, Grad-CAM, Grad-CAM++, and Eigen-CAM. These algorithms also show stability with respect to the threshold used to select tracking targets. In addition, the higher threshold makes the selection robust, and the lower one is useful for analyzing tends of multiple cells in a whole frame efficiently.

Premature babies are admitted to the neonatal intensive care unit (NICU) for several weeks and are generally placed under high medical supervision. The NICU environment is considered to have a bad influence on the formation of the sleep-wake cycle of the neonate, known as the circadian rhythm, because patient monitoring and treatment equipment emit light and noise throughout the day. In order to improve the neonatal environment, researchers have investigated the effect of light and noise on neonates. There are some methods and devices to measure neonatal alertness, but they place on additional burden on neonatal patients or nurses. Therefore, this study proposes an automatic non-contact neonatal alertness state classification method using video images. The proposed method consists of a face region of interest (ROI) location normalization method, histogram of oriented gradients (HOG) and gradient feature-based feature extraction methods, and a neonatal alertness state classification method using machine learning. Comparison experiments using 14 video images of 7 neonatal subjects showed that the weighted support vector machine (w-SVM) using the HOG feature and averaging merge achieved the highest classification performance (micro-F1 of 0.732). In clinical situations, body movement is evaluated primarily to classify waking states. The additional 4 class classification experiments are conducted by combining waking states into a single class, with results that suggest that the proposed facial expression based classification is suitable for the detailed classification of sleeping states.

In a previous work, we proposed an evolutionary computation system designed to solve group decision making multiobjective problems for human groups, which is equivalent to obtaining consensus solutions to multiobjective optimization problems. Multi-human-agent-based evolutionary computation (Mhab-EC) is a primary component of the system, used to obtain converged solutions for multiobjective optimization problems. The other main component is a mechanism that allows owners of simulated human agents to review simulation results thus far and adjust their agents accordingly between successive simulation runs of the Mhab-EC. However, in our previous study, we simply conducted simulations to demonstrate that a single run yielded converged solutions. Consensus solutions were assumed to be obtained through iterations of the Mhab-EC run and agent adjustment. Therefore, in this study, we conducted simulations of the entire system, including the agent adjustment mechanism. For this purpose, we implemented a simple model of agent adjustment by owners to facilitate solution convergence. Simulation results showed that the system indeed yielded converged solutions, which are considered to indicate consensus.

LiDAR SLAM technology is an important method for the accurate navigation of automatic vehicles and is a prerequisite for the safe driving of automatic vehicles in the unstructured road environment of complex parks. This paper proposes a LiDAR fast point cloud registration algorithm that can realize fast and accurate localization and mapping of automatic vehicle point clouds through a combination of a normal distribution transform (NDT) and point-to-line iterative closest point (PLICP). First, the NDT point cloud registration algorithm is applied for the rough registration of point clouds between adjacent frames to achieve a rough estimate of the pose of automatic vehicles. Then, the PLICP point cloud registration algorithm is adopted to correct the rough registration result of the point cloud. This step completes the precise registration of the point cloud and achieves an accurate estimate of the pose of the automatic vehicle. Finally, cloud registration is accumulated over time, and the point cloud information is continuously updated to construct the point cloud map. Through numerous experiments, we compared the proposed algorithm with PLICP. The average number of iterations of the point cloud registration between adjacent frames was reduced by 6.046. The average running time of the point cloud registration between adjacent frames decreased by 43.05156 ms. The efficiency of the point cloud registration calculation increased by approximately 51.7%. By applying the KITTI dataset, the computational efficiency of NDT-ICP was approximately 60% higher than that of LeGO-LOAM. The proposed method realizes the accurate localization and mapping of automatic vehicles relying on vehicle LiDAR in a complex park environment and was applied to a Small Cyclone automatic vehicle. The results indicate that the proposed algorithm is reliable and effective.

Oral mucosal disease is likely to cause various disorders after treatment to occur in a domain called the oral cavity. Therefore, we are developing a diagnostic support system for early screening of oral mucosal disease. There is a problem of individual differences in the cut-out of the disease area from the original intraoral image in system development. In this study, we analyzed the relationships between cutout areas, extracted features and classification rates and investigated the relationship between individual differences. Therefore, we focused on how to eliminate the subjects. Group classification was then performed and identification was performed using an oral mucosal diagnosis support system with ensemble learning. The experimental results revealed relationships between the excision range, identification rate, and feature value.

The performance of conventional torque control for PMSM drive used in electric vehicles (EVs) from the viewpoint of permanent magnet (PM) demagnetization faults has not been satisfactory. Therefore, a combination method based on sliding-mode observer and active disturbance rejection control is presented. First, the model of the PMSM system with PM demagnetization faults is constructed. Then, a sliding-mode observer is designed based on a minimum extended flux linkage to estimate the torque and the PM flux linkages of the system. A current controller is presented based on active disturbance rejection control approach to reject the PM demagnetization faults. The method is useful to improve the control performance of the PMSM drive system. And the system is robust to system parameters variations. Finally, an RT-LAB real-time simulation is used to build a simulation model of hardware-in-the-loop based on the experimentally validated model that is derived from the actual development process for an electric bus. The simulation and experimental results demonstrate the effectiveness of the method.

In this study, we investigate the potential sales forecasts of unhandled bread products in retail stores based on factory shipment data. An embedding-based forecasting method that uses large-scale information network embedding (LINE) and simultaneously considers first- and second-order proximities is developed to define similar neighboring stores using their product–store relationship and to predict their potential sales volume. LINE is a network-embedding method that transforms network data into a low-dimensional distributed representation and requires a low computation time, even when applied to large networks. The results show that our proposed method outperforms a simple prediction method (Baseline) and t-SNE, a well-known dimensionality reduction method for high-dimensional data, in terms of accurate product sales prediction via simulation experiments. Furthermore, we conduct a sensitivity analysis to verify the applicability of our proposed method when the forecasting target is expanded to products sold in fewer stores and in stores with less product variety.

Loop-closure detection (LCD) in large non-stationary environments remains an important challenge in robotic visual simultaneous localization and mapping (vSLAM). To reduce computational and perceptual complexity, it is helpful if a vSLAM system has the ability to perform image change detection (ICD). Unlike previous applications of ICD, time-critical vSLAM applications cannot assume an offline background modeling stage, or rely on maintenance-intensive background models. To address this issue, we introduce a novel maintenance-free ICD framework that requires no background modeling. We demonstrate that LCD can be reused as the main process for ICD with minimal extra cost. Based on these concepts, we develop a novel vSLAM component that enables simultaneous LCD and ICD. ICD experiments based on challenging cross-season LCD scenarios validate the efficacy of the proposed method.

In order to accurately predict blood pressure waveform from pulse waveform, a multiple gated recurrent unit (GRU) model embedded in squeeze-and-excitation network (SENet) is proposed for continuous blood pressure prediction. Firstly, the features of the pulse are extracted from multiple GRU channels. Then, the SENet module is embedded to learn the interdependence among the channels, so as to get the weight of each channel. Finally, the weights were added to each channel and the predicted continuous blood pressure values were obtained by integrating the two linear layers. The experimental results show that the embedded SENet can effectively enhance the predictive ability of multi-GRU structure and obtain good continuous blood pressure waveform. Compared with the LSTM and GRU model without SENet, the MSE errors of the proposed method are reduced by 29.3% and 25.0% respectively, the training time of the proposed method are decreased by 69.8% and 68.7%, the test time is reduced by 65.9% and 25.2% and it has the fewest model parameters.