Soil classification is essential in agriculture, civil engineering, and environmental monitoring, yet conventional methods are labor intensive, time consuming, and highly sensitive to moisture content. This study presents a non-invasive multispectral imaging (MSI) approach for soil type classification, focusing on clay, silt, and sand. A custom built system captured spectral data of clay, silt, and sand samples across the 365–940 nm range under multiple drying stages. Spectral features corresponding to moisture induced color changes were systematically extracted, enabling classification independent of moisture content. Preprocessing steps, including dark current subtraction, cropping, histogram equalization, and dimensionality reduction via Linear Discriminant Analysis (LDA), ensured high quality data. LDA revealed clear separability among soil classes, while Gradient Boosting (GB) classifier achieved the highest accuracy of 93.7% for testing data and 93.3% for validation data. The proposed approach demonstrates strong potential for rapid and scalable applications in precision agriculture, environmental monitoring, and geotechnical analysis.
Deep neural networks (DNNs) offer plenty of challenges in executing efficient computation at edge nodes, primarily due to the huge hardware resource demands. The article proposes HYDRA, hybrid data multiplexing and runtime layer configurable DNN accelerators to overcome the drawbacks. The work proposes a layer-multiplexed approach, which further reuses a single activation function (AF) within the execution of a single layer with improved Fused-Multiply-Accumulate (FMA). The proposed approach works in iterative mode to reuse the same hardware and execute different layers in a configurable fashion. The proposed architectures achieve reductions over 90% of power consumption and resource utilization improvements from state-of-the-art (SOTA) works, with an energy efficiency of 7.04 GOPS/w. The proposed architecture reduces the area-overhead (N-1) times required in bandwidth, AF and layer architecture. The proposed HYDRA architecture supports optimal DNN computations while improving performance on resource-constrained edge devices.
Remote sensing change detection is vital for monitoring environmental and urban transformations but faces challenges like manual feature extraction and sensitivity to noise. Traditional methods and early deep learning models, such as convolutional neural networks (CNNs), struggle to capture long-range dependencies and global context essential for accurate change detection in complex scenes. While Transformer-based models mitigate these issues, their computational complexity limits their applicability in high-resolution remote sensing. Building upon ChangeMamba architecture, which leverages state space models for efficient global context modeling, this paper proposes precision fusion blocks to capture channel-wise temporal variations and per-pixel differences for fine-grained change detection. An enhanced decoder pipeline, incorporating lightweight channel reduction mechanisms, preserves local details with minimal computational cost. Additionally, an optimized loss function combining Cross Entropy, Dice and Lovasz objectives addresses class imbalance and boosts Intersection-over-Union (IoU). Evaluations on SYSU-CD, LEVIR-CD+, and WHU-CD datasets demonstrate superior precision, recall, F1 score, IoU, and overall accuracy compared to state-of-the-art methods, highlighting the approach's robustness for remote sensing change detection. For complete transparency, the codes and pretrained models are accessible at https://github.com/Buddhi19/MambaCD.git
Fetal movement monitoring is an essential part to ensure the safety of a fetus. Counting the number of kicks regularly and keeping track of the pattern is a well known method of evaluating fetal health. This process can be done at home or using a clinical setup but both these methods respectively have limitations such as low accuracy and unavailability to access equipment regularly. Therefore, devices that can be used in a home environment to record fetal movements are considered and algorithms that can identify fetal movements from such recordings are experimented. As attention mechanisms have shown the ability to better identify long time series events, the possibility of better identifying fetal movement occurrence using such models has been discussed in this paper as a novel use case and a novel attempt to correlate statistical parameters with the gestational age have also been discussed so that a fetus with irregularities such as decreased growth can be identified beforehand.
The COVID-19 pandemic has created a shortage of ICU beds and ambulances in Sri Lankan hospitals, leading to overcrowding and patient discomfort. There is a pressing need for diverse data and information to make swift decisions and ensure efficient patient transfers amid the limited availability of ICU beds and ambulances. Ambulances are primarily owned by private hospitals in urban areas, while many hospitals in Sri Lanka lack sufficient ICU beds and facilities. To address these challenges, a proposed mobile platform aims to forecast ICU bed requirements by connecting all government hospitals and ambulances. Powered by AI and utilizing machine learning models, the platform predicts the number of beds needed and records comprehensive ICU-related data. This solution improves the efficiency of medical staff, reduces the risk for critically ill patients, and ultimately lowers the mortality rate, driving the paradigm shift towards achieving Health 4.0 in the healthcare operation model.
Currently, the Rate Splitting Multiple Access (RSMA) scheme is gaining widespread attention as the most promising technique for multiple access in next-generation wireless communications. Its standout feature is the ability to support non-orthogonal transmission, making it increasingly popular for its interference-free transmission capabilities in highly diverse scenarios, which has been a critical bottleneck in existing access schemes. RSMA is considered to outperform its counterparts in terms of achievable rate and overall performance. This paper introduces RSMA and its system model along with its advantages over traditional schemes. It demonstrates the superiority in system performance as compared to orthogonal and non-orthogonal multiple access schemes. Additionally, we highlight the implementation challenges of RSMA and discuss future research opportunities in this direction.
Manual data annotation is an important NLP task but one that takes considerable amount of resources and effort.In spite of the costs, labeling and categorizing entities is essential for NLP tasks such as semantic evaluation.Even though annotation can be done by non-experts in most cases, due to the fact that this requires human labor, the process is costly.Another major challenge encountered in data annotation is maintaining the annotation consistency.Annotation efforts are typically carried out by teams of multiple annotators.The annotations need to maintain the consistency in relation to both the domain truth and annotation format while reducing human errors.Annotating a specialized domain that deviates significantly from the general domain, such as fantasy literature, will see a lot of human error and annotator disagreement.So it is vital that proper guidelines and error reduction mechanisms are enforced.One such way to enforce these constraints is using a specialized application.Such an app can ensure that the notations are consistent, and the labels can be pre-defined or restricted reducing the room for errors.In this paper, we present SHADE, an annotation software that can be used to annotate entities in the high fantasy literature domain.Specifically in Dungeons and Dragons lore extracted from the Forgotten Realms Fandom Wiki.
YouTube is a popular social network for the distribution of videos globally and it is a one of the popular platforms that a large number of people access daily. On the YouTube platform, there are both high-quality videos and low-quality videos. Therefore, it is important to identify high-quality videos among the videos on YouTube to gain a full advantage of them without wasting time and make predictions of the available videos on YouTube. We identified view count, like count, comment count, caption, number of subscribers, tag count, total views, total videos, avg. polarity score, duration secs, title length, and description length as the main factors to decide the video quality. The purpose of this approach is to suggest seven different classification algorithms to make predictions of the quality/integrity of YouTube videos for users. The collected data set was pre-processed as necessary by cleaning the data groups and removing unnecessary attributes by the attribute ranking. This study was conducted using seven classification algorithms Random Forest, Logistic Regression, Support Vector Machine (SVM), Decision Tree, Multilayer Perception (MLP), Naive Bayes, and ensemble learning algorithm that combined the five individual algorithms. The model used 60% training data and 40% testing data for the classification by the Waikato Environment for Knowledge Analysis (WEKA) tool. The accuracy of each algorithm, Precision, Recall, and F-Measure are considered and Random Forest is identified as the best individual algorithm with 96.89% testing accuracy and ensemble learning with 97.21% testing accuracy. Based on the results, this model enables YouTube users to identify the quality/integrity of YouTube videos and take their decisions to view those videos.
This study uses Recursive Feature Elimination (RFE), to select the feature in Exercise Pose Prediction. Firstly, feature selection was performed based on RFE. Among them, eXtreme Gradient Boosting (XGBoost), Random Forest (RF) and Decision Tree (DT) were used as the base estimators for RFE, respectively. The number of selected features was adjusted from 1 to 98, and the step was maintained at 1, to identify the optimal feature combination in the Exercise Pose Prediction. Then, the XGBoost, RF, DT and LR classification algorithms were used to classify and predict the Exercise Pose Prediction features selected by RFE, respectively. Ay last, using the classification results of the classifiers', the performance metrics of the Exercise Pose Prediction were compared by classifiers type. The results show that XGBoost was more suitable for the data after RFE feature selection than RF and DT classifiers. When the base estimator of RFE was XGBoost, the number of selected features was 52, and the performance metrics of the XGBoost classifier were the highest.
The rapid growth in RES (Renewable Energy Sources) penetration of the power systems demands an immense flexibility requirement. To measure the flexibility in power systems, technical flexibility metrics which only account the technical specifications of the generators and operational flexibility metrics which cover the overall operational aspects have been proposed. However, none of these metrics provide any guideline for power system planners to identify the less flexible units in the system and to enhance the system flexibility by upgrading them. This paper proposes a process which utilizes a technical and an operational flexibility metric to meet the aforementioned deficiency. Previously proposed Flexibility Limits Violation Probability (FLVP) metric is used as the operational flexibility metric. A modification for the technical flexibility metric: Normalized Flexibility index (NFI) is presented by emphasizing the necessity of the change and used in the proposed process. Numerical results on a realistic test system are presented to validate the proposed methodology.
This paper presents the design of a battery-free chip-based UHF RFID sensor tag appropriate for temperature measurement inside a poultry egg incubator. The tag employs an EM4325 UHF RFID sensor chip that has a built-in temperature sensor. The antenna was connected to the chip via an internal impedance-matching circuitry and they were etched on an FR4 substrate. Tuning the antenna configuration to match the impedance of the chip and miniaturization of the tag size were performed using parametric optimization. Simulations were performed in ANSYS High-Frequency Structure Simulator (HFSS). The RFID tag, with a compact size of 75 × 20 × 1.6 mm 3 , performs well in terms of impedance matching, bandwidth, and radiation efficiency. Further, the proposed UHF RFID sensor tag demonstrates both far-field and near-field characteristics due to the antenna's dipole topology and loop structure respectively.
Various diseases like brain-related strokes, aphasia and amyotrophic lateral sclerosis (ALS) lead to speech impair-ments. Speech impairments will negatively affect all aspects of the life of an individual and reduce its quality. Speech BCI can provide an alternative communication interface for both healthy and speech-impaired individuals. Imagined speech can be used to develop intuitive Brain Computer Interfaces (BCIs) for subjects with speech impairments. This study focuses on the subject dependency of EEG signals in the context of imagined vowel phoneme classification. Various feature engineering techniques derived from wavelets, statistical analysis, relative wave energy and Hilbert-Huang transforms were studied under both multi-class and binary classifications to compare the results under subject-independent and subject-dependent classifications. Each classification was done using 23 conventional machine learning models with 10-fold cross-validation. The classifications obtained notable accuracies above the chance level for both multi-class and binary classifications. In every scenario, subject-dependent classification had higher accuracies compared to the subject-independent classification. Additionally, the statistical significance of the mean of the results obtained under subject-dependent classification over subject-independent classification was proven under Welch's t-test. Hence, this study provides solid validation that there is a high subject-dependent nature of EEG signals for the task of phoneme imagination.
Endothelial dysfunction (ED) is a reliable biomarker for cardiovascular diseases and thus has significant potential as a screening tool. However, non-invasive methods for measuring endothelial function have a high degree of variability. Triangulation of observations from multiple indicators can mitigate this problem. This work presents Endo-MASS, a modular device capable of simultaneously acquiring multiple biosignals, including photoplethysmography (PPG), impedance plethysmography (IPG), peripheral arterial tonometry (PAT), and temperature. Concurrent measurements are obtained from all four modalities at an adequate sampling rate to measure transient changes in vascular function during reactive hyperemia of the forearm. We confirm that Endo-MASS could differentiate healthy and impaired endothelial function from data collected from a cohort of subjects. Hence the design and implementation of device serve as a foundation for using multiple modalities to improve ED diagnosis, and for research of novel markers of ED such as bioimpedance.
PneuNets (Pneumatic Networks) actuators have gained recent attention in developing actuation systems for Soft Robotic Exoskeletons (SREs), which are serially connected air chambers that are pressurized during actuation. This research aims to investigate the suitability of several popular and widely available silicon rubbers (Silastic RBL-9200-30, Hydrogel, Elastosil 3000–60, Polyethylene Glycol 400, Sylgard184 (Dow corning silicone), and Ecoflex 00–30) to develop a soft robotic rehabilitation glove for flexion movement of the thumb using PneuNets actuators. Followed by a computer simulation study, we found that Silastic RBL-9200-30, Sylgard184 (Dow corning silicone), and Ecoflex 00–30 are best suited to fabricate actuators via 3-D moulding. Compared to 3-D moulds of the same dimensions, Ecoflex 00–30 3-D mould is the best considering low air bubble formation and fixing time of the structure. The soft robotic rehabilitation gloves developed using these three actuators were benchmarked for their flexion angle and actuator pressure and compared against clinical data and reported literature. We report that Ecoflex 00–30 actuators are best suited to develop soft robotic rehabilitation gloves as they agree well with clinical requirements and exhibit comparatively low internal pressures for flexion angle range 30 - 180°.
BERT [Bidirectional Encoder Representations from Transformers] was the first type of neural model which was used hugely with several downstream tasks specially with the generated embedding representation for text classification. But with the existing literature different layer combination was not tested with concatenation as well as mean averaging techniques to represent the features of a text record for short text classification performances. Moreover, it was not properly evaluated with different machine learning algorithms with short text. Therefore, this research work focused on using seven short text type datasets with BERT embedding to identify the best embedding representation and the best machine learning algorithm. Based on conducted experiment, usage of all layers with concatenation techniques was identified as the best embedding representation and bagging as well as support vector machine was identified as the best machine learning algorithm for short text classification. Performance improvements with bagging was more than 6–8% over the other algorithm and concatenation of all layers gave around 5–6 % performance improvements over the other techniques.
In recent years, electric power and energy systems in Japan have faced two challenges: electricity system reform and the large-scale introduction of renewable energy. The electricity system reform included the separation of the generation sector from the transmission and distribution sectors and the full liberalization of retail, including small consumers. In a deregulated environment, several generation companies have created supply-demand schedules, and a power system operator has integrated them into one schedule for the entire power system. Our research group has previously developed methods for daily supply-demand scheduling and operation of the entire grid, including unit commitment (UC), while considering the introduction of a large amount of renewable energy and the utilization of battery storage systems. However, these methods assumed that a power system operator can fully control all the power resources in the grid, but did not consider the electricity markets. Therefore, the present study proposes an estimation method for the supply-demand schedule considering the Japanese spot and balancing markets. In doing so, we improve upon the previously proposed UC method and add additional constraints of reserved power source availability, response speed of the reserved power sources, and duration of the reserved power source to obtain the breakdown of the reserve and regulation capacities that are traded in the Japanese balancing market.
Light emitting diode (LED) lighting systems are widely used in dimming and color-changing applications. When an automated lighting control system is in use, there can be several combinations of dimming levels and colors of lighting fixtures in a given installation. The power consumption of an LED depends on the level of dimming and the color. Even though the lighting system is designed to be balanced over three phases at the full load condition, the system unbalanced conditions are created when the automated lighting control systems are in operation. This paper evaluates the standards applicable to LED systems with control, and phase balancing requirements in lighting systems, followed by an investigation on power quality impact created by mass-scale LED lighting installations in dynamic operation based on several field studies on dimmable LED lighting systems.
Grammatical Error Correction (GEC) is crucial for improving the readability and comprehension of text. Although substantial advancements have been achieved in this area for widely-spoken languages such as English, the focus on the development of GEC tools for less common languages such as Sinhala has been inadequate. Sinhala is a language spoken by more than 16 million people in Sri Lanka, known for its rich morphology, and complex grammar structures that pose a challenge for Sinhala GEC systems. This paper presents a novel GEC approach that utilizes Google machine translation, cross-linguistic knowledge and rule-based techniques augmented by machine learning to analyze complex Sinhala sentences. We focus on analyzing Sinhala verb agreement rules, and object validation rules in Sinhala active voice sentences. Additionally, we address the major challenges in Sinhala GEC, such as subject and object detection and the detection of grammatical features of nouns, including animacy, gender, and number. Our findings indicate that the GEC methodology presented achieved an accuracy of 75.61 %. Additionally, the gender and number detection components produced an accuracy of 90.89% and 92.33%, respectively. These results demonstrate the effectiveness of our approach in identifying and correcting errors in complex Sinhala sentences. Our approach is particularly useful in languages with rich morphology and limited annotated data.
In this paper, we consider a spatial modulation aided indoor visible light communication system in which an intelligent reflective surface (IRS) is used to assist communication. The orientation of the individual mirror elements of the IRS can be changed to achieve a higher performance. In particular, we formulate an optimization problem to find the orientation angles of each mirror element that minimizes the bit error rate (BER). The original problem is transformed into a reflected spot finding problem and several solutions that show good performance are developed. Moreover, we propose IRS pattern design to convey extra information bits. Our results show that the presented genetic algorithm based area assignment achieves a better BER performance as compared to conventional scheme without IRS in more than one-quarter of the floor area.
Continuous measurement of harmonic levels in a power system is an important task to maintain the power quality of the network. There are online power quality and harmonic measuring devices installed at various locations in the transmission network of Sri Lanka. Recent measurements from such devices show violations of some of the individual voltage harmonics as well as total harmonics distortion (THD) at specific times of the day. These violations are significant around the newly constructed large-scale wind farm in the Northern region of the island. Harmonic violations can occur due to various reasons. This study discusses the possible reasons for this harmonic amplification and identifies the most influential factors. In addition, several means to mitigate the voltage harmonic violation in the identified area are discussed based on the DIgSILENT PowerFactory simulations.
