Correct Classification Accuracy Research Articles

Identifying neuroimaging biomarkers in major depressive disorder using machine learning algorithms and functional near-infrared spectroscopy (fNIRS) during verbal fluency task

One of the most prevalent psychiatric disorders is major depressive disorder (MDD), which increases the probability of suicidal ideation or untimely demise. Abnormal frontal hemodynamic changes detected by functional near-infrared spectroscopy (fNIRS) during verbal fluency task (VFT) have the potential to be used as an objective indicator for assessing clinical symptoms. However, comprehensive quantitative and objective assessment instruments for individuals who exhibit symptoms suggestive of depression remain undeveloped. Drawing from a total of 467 samples in a large-scale dataset comprising 289 MDD patients and 178 healthy controls, fNIRS measurements were obtained throughout the VFT. To identify unique MDD biomarkers, this research introduced a data representation approach for extracting spatiotemporal features from fNIRS signals, which were subsequently utilized as potential predictors. Machine learning classifiers (e.g., Gradient Boosted Decision Trees (GBDT) and Multilayer Perceptron) were implemented to assess the ability to predict selected features. The mean and standard deviation of the cross-validation indicated that the GBDT model, when combined with the 180-feature pattern, distinguishes patients with MDD from healthy controls in the most effective manner. The accuracy of correct classification for the test set was 0.829 ± 0.053, with an AUC of 0.895 (95 % CI: 0.864–0.925) and a sensitivity of 0.914 ± 0.051. Channels that made the most important contribution to the identification of MDD were identified using Shapley Additive Explanations method, located in the frontopolar area and the dorsolateral prefrontal cortex, as well as pars triangularis Broca's area. Assessment of abnormal prefrontal activity during the VFT in MDD serves as an objectively measurable biomarker that could be utilized to evaluate cognitive deficits and facilitate early screening for MDD. The model suggested in this research could be applied to large-scale case-control fNIRS datasets to detect unique characteristics of MDD and offer clinicians an objective biomarker-based analytical instrument to assist in the evaluation of suspicious cases.

Off-line handwritten signature verification based on machine learning

This paper describes the results of recognizing handwritten signatures. For the experiments, the database of handwritten signatures BHSig260-Bengali, BHSig260-Hindi, CEDAR and TUIT was used. For classification, four options were used to reduce the signatures to sizes: 200×120, 250×150, 300×150 and 400×200 pixels. These images served as input for the proposed network architecture. As a result of testing the proposed approach, the average accuracy of correct classification of signatures on images of size 250×150 was achieved: for the CEDAR database it was 94.38%, for the BHSig260-Hindi database it was 95.63%, for the BHSig260-Bengali database it was 97.50% and for TUIT base is 90.04%.

Eucalyptus Species Discrimination Using Hyperspectral Sensor Data and Machine Learning

The identification of tree species is very useful for the management and monitoring of forest resources. When paired with machine learning (ML) algorithms, species identification based on spectral bands from a hyperspectral sensor can contribute to developing technologies that enable accurate forest inventories to be completed efficiently, reducing labor and time. This is the first study to evaluate the effectiveness of classification of five eucalyptus species (E. camaldulensis, Corymbia citriodora, E. saligna, E. grandis, and E. urophyla) using hyperspectral images and machine learning. Spectral readings were taken from 200 leaves of each species and divided into three dataset sizes: one set containing 50 samples per species, a second with 100 samples per species, and a third set with 200 samples per species. The ML algorithms tested were multilayer perceptron artificial neural network (ANN), decision trees (J48 and REPTree algorithms), and random forest (RF). As a control, a conventional approach by logistic regression (LR) was used. Eucalyptus species were classified by ML algorithms using a randomized stratified cross-validation with 10 folds. After obtaining the percentage of correct classification (CC) and F-measure accuracy metrics, the means were grouped by the Scott–Knott test at 5% probability. Our findings revealed the existence of distinct spectral curves between the species, with the differences being more marked from the 700 nm range onwards. The most accurate ML algorithm for identifying eucalyptus species was ANN. There was no statistical difference for CC between the three dataset sizes. Therefore, it was determined that 50 leaves would be sufficient to accurately differentiate the eucalyptus species evaluated. Our study represents an important scientific advance for forest inventories and breeding programs with applications in both forest plantations and native forest areas as it proposes a fast, accurate, and large-scale species-level classification approach.

Sybil Attack Detection in Internet of Flying Things-IoFT: A Machine Learning Approach

Sybil attack refers to the situation when a malicious node falsely claims to have numerous identities and is known to be one of the security threats to the Internet of Things (IoT). Due to recent increase usage of unmanned aerial vehicles (UAVs) in various applications, Sybil attack has been identified as a threat to the flying ad hoc network (FANET) paradigm and its integration with the IoT to form the Internet of Flying Things (IoFT). In this paper, we propose an intelligent Sybil attack detection approach for FANETs-based IoFT using physical layer characteristics of the radio signals emitted from the UAVs as detected by two ground nodes. A supervised machine learning approach is employed and experimented with several different classifiers available in the Weka workbench platform. The experiment was carried out based on two features of the radio signals, namely, the received signal strength difference (RSSD) and the time difference of arrival (TDoA). Simulation results revealed that the proposed scheme can achieve a high correct classification accuracy of above 91% on average, even for smart malicious nodes with power control capability operating at power levels not directly trained. In addition to its high performance, the proposed scheme is also less susceptible to various attacks commonly carried out on the upper layers, such as data spoofing, due to the use of only intrinsically generated physical layer data. Furthermore, no additional communications overheads of the UAV nodes are required for the functionality of this scheme.

Machine Learning Based Performance Analysis of Video Object Detection and Classification Using Modified Yolov3 and Mobilenet Algorithm

Detecting foreground objects in video is crucial in various machine vision applications and computerized video surveillance technologies. Object tracking and detection are essential in object identification, surveillance, and navigation approaches. Object detection is the technique of differentiating between background and foreground features in a photograph. Recent improvements in vision systems, including distributed smart cameras, have inspired researchers to develop enhanced machine vision applications for embedded systems. The efficiency of featured object detection algorithms declines as dynamic video data increases as contrasted to conventional object detection methods. Moving subjects that are blurred, fast-moving objects, backdrop occlusion, or dynamic background shifts within the foreground area of a video frame can all cause problems. These challenges result in insufficient prominence detection. This work develops a deep-learning model to overcome this issue. For object detection, a novel method utilizing YOLOv3 and MobileNet was built. First, rather than picking predefined feature maps in the conventional YOLOv3 architecture, the technique for determining feature maps in the MobileNet is optimized based on examining the receptive fields. This work focuses on three primary processes: object detection, recognition, and classification, to classify moving objects before shared features. Compared to existing algorithms, experimental findings on public datasets and our dataset reveal that the suggested approach achieves 99% correct classification accuracy for urban settings with moving objects. Experiments reveal that the suggested model beats existing cutting-edge models by speed and computation.

Determination of beef tenderness based on airflow pressure combined with structural light three-dimensional (3D) vision technology

The main factor affecting beef quality, consumer satisfaction, and purchase decisions is beef tenderness. In this study, a rapid nondestructive testing method for beef tenderness based on airflow pressure combined with structural light 3D vision technology was proposed. The structural light 3D camera was used to scan the 3D point cloud deformation information of the beef surface after the airflow acted on it for 1.8 s. Six deformation characteristics and three point cloud characteristics of the beef surface depression region were obtained by using denoising, point cloud rotation, point cloud segmentation, point cloud descending sampling, alphaShape, and other algorithms. A total of nine characteristics were mainly concentrated in the first five principal components (PCs). Therefore, the first five PCs were put into three different models. The results showed that the Extreme Learning Machine (ELM) model had a comparatively higher prediction effect for the prediction of beef shear force, with a root mean square error of prediction (RMSEP) of 11.1389 and a correlation coefficient (R) of 0.8356. In addition, the correct classification accuracy of the ELM model for tender beef achieved 92.96%. The overall classification accuracy reached 93.33%. Consequently, the proposed methods and technology can be applied for beef tenderness detection.

Ship detection and classification based on cascaded detection of hull and wake from optical satellite remote sensing imagery

ABSTRACT Satellite remote-sensing provides a cost- and time-effective tool for ship monitoring at sea. Most existing approaches focused on extraction of ship locations using either hull or wake. In this paper, a method of cascaded detection of ship hull and wake was proposed to locate and classify ships using high-resolution satellite imagery. Candidate hulls were fast located through phase spectrum of Fourier transform. A hull refining module was then executed to acquire accurate shapes of candidate hull. False alarms were removed through the shape features and textures of candidate hulls. The probability that a candidate hull is determined as a real one increased with the presence of wakes. After true ships were determined, ship classification was conducted using a fuzzy classifier combining both hull and wake information. The proposed method was implemented to Gaofen-1 panchromatic and multispectral (PMS) imagery and showed good performance for ship detection with recall, precision, overall accuracy, and specificity of 90.1%, 88.1%, 98.8%, and 99.3%, respectively, better than other state-of-the-art coarse-to-fine ship detection methods. Ship classification was successfully achieved for ships with detected wakes. The accuracy of correct classification was 83.8% while the proportion of false classification was 1.0%. Factors influencing the accuracy of the developed method, including texture features and classifiers combination and key parameters of the method, were also discussed.

Reconstructing habitat use, identifying origin and discrimination of the barramundi (wild and farmed) populations using otolith stable isotope analysis

Barramundi or Asian seabass, Lates calcarifer (Bloch, 1790), is a euryhaline fish that is widely distributed in the Indo-West Pacific. This species is a commercially important fish which is extensively farmed in Taiwan. Currently, there is a lack of information on the origin, habitat use and migratory behavior of barramundi in Taiwanese waters. Here, δ13C and δ18O values of otolith growth time series were evaluated in barramundi (wild and farmed), and salinity was estimated from otolith δ18O values to report the habitat use and origin of wild individuals. Also, wild individuals and farmed individuals were discriminated using δ13C and δ18O values of otoliths to explore whether the individuals captured from Danshui River Estuary have aquaculture origin. The wild individuals collected from the Danshui River Estuary displayed a relatively wide range of δ13C and δ18O values however farmed samples showed a consistent change among the individuals within a group. Based on otolith δ18O values and estimated salinity, the most of wild fish samples were born in the brackish water environment. The estimated salinity experienced by the wild individuals also suggests that migration behavior varies among individuals. Moreover, some wild individuals showed a regular migration pattern but the range of salinity experienced by individuals varied greatly. By comparing the δ13C and δ18O values, it was found that significant differences existed among groups of fish and overall the correct classification accuracy was 95% using quadratic discriminant analysis. The δ13C and δ18O values of the otolith core revealed that only one sample caught in the Danshui River Estuary might have aquaculture origin. In this study, stable isotope (δ13C and δ18O) values were effectively applied to reconstruct the life history of barramundi and to distinguish individuals having a different origin. The results of this study can be used as a reference for sustainable management of this important fishery resource.

Hyperspectral Imaging and Chemometrics for Authentication of Extra Virgin Olive Oil: A Comparative Approach with FTIR, UV-VIS, Raman, and GC-MS.

Limited information on monitoring adulteration in extra virgin olive oil (EVOO) by hyperspectral imaging (HSI) exists. This work presents a comparative study of chemometrics for the authentication and quantification of adulteration in EVOO with cheaper edible oils using GC-MS, HSI, FTIR, Raman and UV-Vis spectroscopies. The adulteration mixtures were prepared by separately blending safflower oil, corn oil, soybean oil, canola oil, sunflower oil, and sesame oil with authentic EVOO in different concentrations (0-20%, m/m). Partial least squares-discriminant analysis (PLS-DA) and PLS regression models were then built for the classification and quantification of adulteration in olive oil, respectively. HSI, FTIR, UV-Vis, Raman, and GC-MS combined with PLS-DA achieved correct classification accuracies of 100%, 99.8%, 99.6%, 96.6%, and 93.7%, respectively, in the discrimination of authentic and adulterated olive oil. The overall PLS regression model using HSI data was the best in predicting the concentration of adulterants in olive oil with a low root mean square error of prediction (RMSEP) of 1.1%, high R2pred (0.97), and high residual predictive deviation (RPD) of 6.0. The findings suggest the potential of HSI technology as a fast and non-destructive technique to control fraud in the olive oil industry.

Gastrointestinal Diseases Classification Using Deep Transfer Learning and Features Optimization

Gastrointestinal diseases like ulcers, polyps’, and bleeding are increasing rapidly in the world over the last decade. On average 0.7 million cases are reported worldwide every year. The main cause of gastrointestinal diseases is a Helicobacter Pylori (H. Pylori) bacterium that presents in more than 50% of people around the globe. Many researchers have proposed different methods for gastrointestinal disease using computer vision techniques. Few of them focused on the detection process and the rest of them performed classification. The major challenges that they faced are the similarity of infected and healthy regions that misleads the correct classification accuracy. In this work, we proposed a technique based on Mask Recurrent-Convolutional Neural Network (R-CNN) and fine-tuned pre-trained ResNet-50 and ResNet-152 networks for feature extraction. Initially, the region of interest is detected using Mask R-CNN which is later utilized for the training of fine-tuned models through transfer learning. Features are extracted from fine-tuned models that are later fused using a serial approach. Moreover, an Improved Ant Colony Optimization (ACO) algorithm has also opted for the best feature selection from the fused feature vector. The best-selected features are finally classified using machine learning techniques. The experimental process was conducted on the publicly available dataset and obtained an improved accuracy of 96.43%. In comparison with state-of-the-art techniques, it is observed that the proposed accuracy is improved.

Simultaneous 18F-FDG PET/MR metabolic and structural changes in visual snow syndrome and diagnostic use

PurposeVisual snow syndrome (VSS) is a recently recognized chronic neurologic condition characterized by the constant perceiving of tiny flickering dots throughout the entire visual field. Metabolic overactivity and grey matter volume increase in the lingual gyrus has been reported. We investigated this by 18F-FDG PET/MR in comparison to healthy controls. Aside from voxel-based characterization, the classification accuracy of volume-of-interest (VOI)-based multimodal assessment was evaluated, also in comparison with visual analysis.MethodsSimultaneous 18F-FDG PET and MR imaging was performed in 7 patients with VSS (24.6 ± 5.7 years; 5 M/2F) and 15 age-matched healthy controls (CON) (28.0 ± 5.3 years; 8 M/7F). SPM12 and voxel-based morphometric analysis was performed. A VOI-based discriminant analysis was performed with relative 18F-FDG uptake, MR grey matter (GM) volumes and their combination. A visual analysis was done by two blinded experienced readers.ResultsRelative increased hypermetabolism was found in VSS patients in the lingual gyrus and cuneus (pFWE < 0.05, peak change + 24%), and hypometabolism in the mesiotemporal cortex (pheight,uncorr < 0.001, peak change − 14%). VSS patients also had increased GM volume in the limbic system and frontotemporal cortex bilaterally (pFWE < 0.05), and in the left secondary and associative visual cortex and in the left lingual gyrus (pheight,uncorr < 0.001). Discriminant analysis resulted in 100% correct classification accuracy for 18F-FDG with lingual gyrus, cuneus and lateral occipital lobe (BA 17 and BA 18) as main discriminators. Unimodal MR- and combined 18F-FDG + MR classification resulted in an accuracy of 91% and 95%, respectively. Visual analysis of 18F-FDG was highly observer dependent.ConclusionPatients with VSS have highly significant structural and metabolic abnormalities in the visual and limbic system. VOI-based discriminant analysis of 18F-FDG PET allows reliable individual classification versus controls, whereas visual analysis of experienced observers was highly variable. Further investigation in larger series, also in comparison to VSS mimicking disorders such as migraine, is warranted.Trail registration: Retrospectively registered at clinicaltrials.gov under NCT05569733 on Oct 5, 2022.

Classification of breast cancer recurrence based on imputed data: a simulation study

Several studies have been conducted to classify various real life events but few are in medical fields; particularly about breast recurrence under statistical techniques. To our knowledge, there is no reported comparison of statistical classification accuracy and classifiers’ discriminative ability on breast cancer recurrence in presence of imputed missing data. Therefore, this article aims to fill this analysis gap by comparing the performance of binary classifiers (logistic regression, linear and quadratic discriminant analysis) using several datasets resulted from imputation process using various simulation conditions. Our study aids the knowledge about how classifiers’ accuracy and discriminative ability in classifying a binary outcome variable are affected by the presence of imputed numerical missing data. We simulated incomplete datasets with 15, 30, 45 and 60% of missingness under Missing At Random (MAR) and Missing Completely At Random (MCAR) mechanisms. Mean imputation, hot deck, k-nearest neighbour, multiple imputations via chained equation, expected-maximisation, and predictive mean matching were used to impute incomplete datasets. For each classifier, correct classification accuracy and area under the Receiver Operating Characteristic (ROC) curves under MAR and MCAR mechanisms were compared. The linear discriminant classifier attained the highest classification accuracy (73.9%) based on mean-imputed data at 45% of missing data under MCAR mechanism. As a classifier, the logistic regression based on predictive mean matching imputed-data yields the greatest areas under ROC curves (0.6418) at 30% missingness while k-nearest neighbour tops the value (0.6428) at 60% of missing data under MCAR mechanism.

Fourier transform infrared (FTIR) spectroscopy approach combined with discriminant analysis and prediction model for crude palm oil authentication of different geographical and temporal origins

The traceability of palm oil to determine its geographical origins within the palm oil supply chains is of importance to ensure quality and sustainable production. The current traceability systems are insufficient, thus requiring a more viable and effective approach to improve the traceability framework of the palm oil supply chain. In this study, the feasibility of Fourier transform infrared (FTIR) and chemometric methods for the traceability of crude palm oil (CPO) samples in Malaysia was investigated. The orthogonal partial least squares-discriminant analysis (OPLS-DA) was used to examine the FTIR spectral data of 354 CPO samples collected from four different mills in Malaysia from February until July 2019. The optimal OPLS-DA models were achieved with standard normal variate (SNV) preprocessing and Pareto (Par) scaling to discriminate the CPO according to its geographical provenance and time of sampling. OPLS-DA models using CPO fingerprints successfully discriminated CPO samples according to their geographical origin and production month, with a total accuracy of correct classification for the blind samples (test set) in the prediction models greater than 90% and with the area under the receiver operating characteristic (ROC) curve values exceeding 0.9. The OPLS-DA loading plot indicated that spectral regions associated with free fatty acid (FFA), iodine value (IV) as well as the saturated and unsaturated fatty acid content were responsible for the discrimination patterns and can be potentially used as spectral markers to differentiate CPO samples. Overall, the study shows that combination of FTIR fingerprinting and chemometric methods could serve as an effective tool to trace and discriminating CPO samples according to their geographical origins and production times. The approaches used in the study present an innovative strategy for monitoring quality of palm oil, and moving forward will also facilitate efforts at successful tracking of palm oil across supply chains, especially when combined with additional analytical techniques.

Non-targeted analysis of VOCs by HS-SPME-G C/MS coupled with chemometrics as a potential tool for authentication of White Kołuda oat goose

Abstract This study tested the possibility of using non-targeted analysis of volatile organic compounds by headspace solid-phase microextraction-gas chromatography-mass spectrometry coupled with chemometrics as a potential tool for differentiating leg meat of oat- and wheat-fed (ad libitum) White Kołuda geese. Thirty-six classification models were obtained for which the correct classification rate and classification accuracy for oatfed and wheat-fed geese were calculated based on a seven-fold cross-validation. Generally, the most advantageous method of the sample preparation was the high-temperature heat treatment version, whereas the highest correct classification rate was obtained when the chemometric analysis was carried out on the female, then male, and finally male + female variant of group comparisons (P&lt;0.01). Furthermore, log-transformation appeared to be a slightly better data preprocessing technique in comparison to systematic ratio normalization. The obtained classification models can potentially differentiate the meat of oat-fattened from wheat-fattened White Kołuda geese.

Face morphing attacks and face image quality: The effect of morphing and the unsupervised attack detection by quality

Morphing attacks are a form of presentation attacks that gathered increasing attention in recent years. A morphed image can be successfully verified to multiple identities. This operation, therefore, poses serious security issues related to the ability of a travel or identity document to be verified to belong to multiple persons. Previous studies touched on the issue of the quality of morphing attack images, however, with the main goal of quantitatively proofing the realistic appearance of the produced morphing attacks. The authors theorise that the morphing processes might have an effect on both, the perceptual image quality and the image utility in face recognition (FR) when compared to bona fide samples. Towards investigating this theory, this work provides an extensive analysis of the effect of morphing on face image quality, including both general image quality measures and face image utility measures. This analysis is not limited to a single morphing technique but rather looks at six different morphing techniques and five different data sources using ten different quality measures. This analysis reveals consistent separability between the quality scores of morphing attack and bona fide samples measured by certain quality measures. The authors’ study goes further to build on this effect and investigate the possibility of performing unsupervised morphing attack detection (MAD) based on quality scores. The authors’ study looks into intra- and inter-dataset detectability to evaluate the generalisability of such a detection concept on different morphing techniques and bona fide sources. The authors’ final results point out that a set of quality measures, such as MagFace and CNNIQA, can be used to perform unsupervised and generalised MAD with a correct classification accuracy of over 70%.

Mathematical Model Simulation of Detailed Classification of Telemedicine Sensing Data

Medical and health field is a hot application field of wireless sensor networks. How to correctly refine and classify telemedicine sensor data is the research focus in related fields. Therefore, a detailed classification mathematical model simulation of telemedicine sensor data based on multi feature fusion is proposed. On the basis of telemedicine sensor data acquisition, it is preprocessed to reduce the computational overhead of detailed classification. The reliability features of the preprocessed telemedicine sensing data are extracted, the extracted features are fused by the principal component analysis method, and the refined classification model of telemedicine sensing data is constructed based on the principle of machine learning. The fused features are input into the model to complete the refined classification of telemedicine sensing data. The experimental results show that the correct refinement classification rate of the proposed method is more than 90%, the refinement classification accuracy is higher than 98.5%, the convergence speed is good, and the refinement classification time is 4 ~ 12 s, which proves that the correct refinement classification rate and accuracy of the proposed method are high, the classification time is short, and has good application performance.

COMPARATIVE PERFORMANCE EVALUATION OF EFFICIENCY FOR HIGH DIMENSIONAL CLASSIFICATION METHODS

This paper aimed to determine the efficiency of classifiers for high-dimensional classification methods. It also investigated whether an extreme minimum misclassification rate translates into robust efficiency. To ensure an acceptable procedure, a benchmark evaluation threshold (BETH) was proposed as a metric to analyze the comparative performance for high-dimensional classification methods. A simplified performance metric was derived to show the efficiency of different classification methods. To achieve the objectives, the existing probability of correct classification (PCC) or classification accuracy reported in five different articles was used to generate the BETH value. Then, a comparative analysis was performed between the application of BETH value and the well-established PCC value ,derived from the confusion matrix. The analysis indicated that the BETH procedure had a minimum misclassification rate, unlike the Optimal method. The results also revealed that as the PCC inclined toward unity value, the misclassification rate between the two methods (BETH and PCC) became extremely irrelevant. The study revealed that the BETH method was invariant to the performance established by the classifiers using the PCC criterion but demonstrated more relevant aspects of robustness and minimum misclassification rate as compared to the PCC method. In addition, the comparative analysis affirmed that the BETH method exhibited more robust efficiency than the Optimal method. The study concluded that a minimum misclassification rate yields robust performance efficiency.

EVALUATION OF MACHINE LEARNING MODELS TO IDENTIFY PEACH VARIETIES BASED ON LEAF COLOR

Machine learning and deep learning approaches are applied in different areas of the agricultural sector, particularly in the digital image-based identification of characteristics of interest in crops. In this research, the performance of three machine learning classifiers was evaluated: support vector machine (SVM), random forest (RF), and multilayer perceptron (MLP). The aim was to identify four varieties of peach (Prunus persica L. Batsch) (CP-03-06, Oro Azteca, Oro San Juan, and Cardenal), based on the color of digital images of the upper and lower side of leaves, represented by two color spaces: RGB (red, green, blue) and HSV (hue, saturation, value). The classifiers were trained and evaluated based on six data input scenarios, defined by the combinations of the upper, lower, and both sides of the leaf with the RGB and HSV color spaces. The three machine learning classifiers (SVM, RF, and MLP) achieved their best prediction performance when they examined the color characteristics of the upper side of leaves transformed to the HSV color space. The SVM classifier outperformed RF and MLP. SVM achieved a global average correct classification accuracy of 84.1 %, F1 macro of 83.7 %, and area under the ROC curve (AUC macro) of 0.93. The Oro Azteca variety reached the highest classification rate with a F1 score of 87.9 % and the Oro San Juan variety obtained the lowest rate with a F1 of 71.3 %.

Sensing System for Plegic or Paretic Hands Self-Training Motivation.

Patients after stroke with paretic or plegic hands require frequent exercises to promote neuroplasticity and to improve hand joint mobilization. Available devices for hand exercising are intended for persons with some level of hand control or provide continuous passive motion with limited patient involvement. Patients can benefit from self-exercising where they use the other hand to exercise the plegic or paretic one. However, post-stroke neuropsychological complications, apathy, and cognitive impairments such as forgetfulness make regular self-exercising difficult. This paper describes Przypominajka v2—a system intended to support self-exercising, remind about it, and motivate patients. We propose a glove-based device with an on-device machine-learning-based exercise scoring, a tablet-based interface, and a web-based application for therapists. The feasibility of on-device inference and the accuracy of correct exercise classification was evaluated on four healthy participants. Whole system use was described in a case study with a patient with a paretic hand. The anomaly classification has an accuracy of 91.3% and value of 91.6% but achieves poorer results for new users (78% and 81%). The case study showed that patients had a positive reaction to exercising with Przypominajka, but there were issues relating to sensor glove: ease of putting on and clarity of instructions. The paper presents a new way in which sensor systems can support the rehabilitation of after-stroke patients with an on-device machine-learning-based classification that can accurately score and contribute to patient motivation.

Micro-Motion Classification of Flying Bird and Rotor Drones via Data Augmentation and Modified Multi-Scale CNN

Aiming at the difficult problem of the classification between flying bird and rotary-wing drone by radar, a micro-motion feature classification method is proposed in this paper. Using K-band frequency modulated continuous wave (FMCW) radar, data acquisition of five types of rotor drones (SJRC S70 W, DJI Mavic Air 2, DJI Inspire 2, hexacopter, and single-propeller fixed-wing drone) and flying birds is carried out under indoor and outdoor scenes. Then, the feature extraction and parameterization of the corresponding micro-Doppler (m-D) signal are performed using time-frequency (T-F) analysis. In order to increase the number of effective datasets and enhance m-D features, the data augmentation method is designed by setting the amplitude scope displayed in T-F graph and adopting feature fusion of the range-time (modulation periods) graph and T-F graph. A multi-scale convolutional neural network (CNN) is employed and modified, which can extract both the global and local information of the target’s m-D features and reduce the parameter calculation burden. Validation with the measured dataset of different targets using FMCW radar shows that the average correct classification accuracy of drones and flying birds for short and long range experiments of the proposed algorithm is 9.4% and 4.6% higher than the Alexnet- and VGG16-based CNN methods, respectively.