Search

Abstract This paper introduces an enhanced object detection algorithm tailored for indoor shopping applications, a critical component of smart cities and smart shopping ecosystems. The proposed method builds on the YOLOv8n algorithm by integrating a ParNetAttention module into the backbone's C2f module, creating the novel C2f‐ParNet structure. This innovation enhances feature extraction, crucial for detecting intricate details in complex indoor environments. Additionally, the channel‐wise attention‐recurrent feature extraction (CARAFE) module is incorporated into the neck network, improving target feature fusion and focus on objects of interest, thereby boosting detection accuracy. To optimize training efficiency, the model employs the Wise Intersection over Union (WIoUv3) as its regression loss function, accelerating data convergence and improving performance. Experimental results demonstrate the enhanced YOLOv8n achieves a mean average precision (mAP) at 50% threshold (mAP@50) of 61.2%, a 1.2 percentage point improvement over the baseline. The fully optimized algorithm achieves an mAP@50 of 65.9% and an F1 score of 63.5%, outperforming both the original YOLOv8n and existing algorithms. Furthermore, with a frame rate of 106.5 FPS and computational complexity of just 12.9 GFLOPs (Giga Floating‐Point Operations per Second), this approach balances high performance with lightweight efficiency, making it ideal for real‐time applications in smart retail environments.

Summary: Achieving lightweight real-time object detection necessitates balancing model compression with detection accuracy, a difficulty exacerbated by low redundancy and uneven contributions from convolutional layers. As an alternative to traditional methods, we propose Rigorous Gradation Pruning (RGP), which uses a desensitized first-order Taylor approximation to assess filter importance, enabling precise pruning of redundant kernels. This approach includes the iterative reassessment of layer significance to protect essential layers, ensuring effective detection performance. We applied RGP to YOLOv8 detectors and tested it on GTSDB, Seaships, and COCO datasets. On GTSDB, RGP achieved 80% compression of YOLOv8n with only a 0.11% drop in mAP0.5, while increasing frames per second (FPS) by 43.84%. For YOLOv8x, RGP achieved 90% compression, a 1.26% mAP0.5:0.95 increase, and a 112.66% FPS boost. Significant compression was also achieved on Seaships and COCO datasets, demonstrating RGP’s robustness across diverse object detection tasks and its potential for advancing efficient, high-speed detection models.

Abstract Maritime objects frequently exhibit low-quality and insufficient feature information, particularly in complex maritime environments characterized by challenges such as small objects, waves, and reflections. This situation poses significant challenges to the development of reliable object detection including the strategies of loss function and the feature understanding capabilities in common YOLOv8 (You Only Look Once) detectors. Furthermore, the widespread adoption and unmanned operation of intelligent ships have generated increasing demands on the computational efficiency and cost of object detection hardware, necessitating the development of more lightweight network architectures. This study proposes the EL-YOLO (Efficient Lightweight You Only Look Once) algorithm based on YOLOv8, designed specifically for intelligent ship object detection. EL-YOLO incorporates novel features, including adequate wise IoU (AWIoU) for improved bounding box regression, shortcut multi-fuse neck (SMFN) for a comprehensive analysis of features, and greedy-driven filter pruning (GDFP) to achieve a streamlined and lightweight network design. The findings of this study demonstrate notable advancements in both detection accuracy and lightweight characteristics across diverse maritime scenarios. EL-YOLO exhibits superior performance in intelligent ship object detection using RGB cameras, showcasing a significant improvement compared to standard YOLOv8 models.

Abstract The article presents a comprehensive summary of Intelligent Ship Marine Object Detection (ISMOD) based on the RGB Camera. Marine object detection plays a pivotal role in enabling intelligent ships to acquire crucial data and security assurances for autonomous navigation. Among the various detection sensors, the RGB Camera is an informative and cost‐effective tool with a wide range of civil applications. In the beginning, the ISMOD metrics based on the RGB camera is analyzed from three significant aspects, namely accuracy, speed, and robustness. Subsequently, the latest research status and comparative overview are presented, encompassing three mainstream detection methods: traditional detection, deep learning detection, and sensor fusion detection. Finally, the existing challenges of ISMOD are discussed and future development trends are recommended. The results demonstrate that forthcoming development will predominantly concentrate on deep learning approaches, complemented by other techniques. It is imperative to advance detection performance in domains such as deep fusion, multi‐feature extraction, multi‐fusion technology, and lightweight detection architecture.

Near infrared spectroscopy (NIRS) is a widely used analytical technique for non-destructive analysis of various materials including food fraud detection. However, the accurate calibration of NIRS data can be challenging due to the complexity of the underlying relationships between the spectral data and the target variables of interest. Ensemble learning, which combines multiple models to make predictions, has been shown to improve the accuracy and robustness of predictive models in various domains. This paper proposes stacking ensemble machine learning (SEML) for calibration of NIRS data with two levels of learning involved. Eight (8) spectroscopy datasets from public repository and previously published works by the authors are used as the case study. The model well generalized the data in the respective regression tasks with of at least »0.8 in the test samples and in the respective classification tasks with classification accuracy (CA) of at least »0.8 also. In addition, the proposed SEML can improve, or at least reach par with, the accuracy of individual base learners in both train and test samples for all cases of regression and classification datasets. It shows superior performance in test samples for both regression and classification datasets with respectively ranging from 0.86 to nearly 1 and CA ranging from 0.89 to 1. ABSTRAK: Spektroskopi inframerah dekat (NIRS) adalah teknik analitikal yang banyak digunakan bagi analisa pelbagai bahan tanpa merosakkan bahan termasuk ketika mengesan penipuan makanan. Walau bagaimanapun, kalibrasi yang tepat bagi data NIRS adalah sangat mencabar kerana hubungan antara data spektral dan pemboleh ubah sasaran yang ingin dikaji bersifat kompleks. Gabungan pembelajaran (Ensemble learning), iaitu gabungan pelbagai model bagi membuat prediksi, telah terbukti dapat meningkatkan ketepatan dan kecekapan model prediksi dalam pelbagai bentuk. Kajian ini mencadangkan Turutan Gabungan Pembelajaran Mesin (Stacking Ensemble Machine Learning ) (SEML), bagi teknik penentu ukuran data NIRS melibatkan dua tahap pembelajaran. Lapan (8) set data spektroskopi dari repositori awam dan kajian terdahulu oleh pengarang telah digunakan sebagai kes kajian. Model ini menggeneralisasi data dalam tugas regresi masing-masing sebanyak ?0.8 bagi sampel ujian dan pengelasan tugas masing-masing dengan ketepatan klasifikasi (CA) sekurang-kurangnya ?0.8. Tambahan, SEML yang dicadangkan ini dapat membantu, atau sekurang-kurangnya setanding dengan ketepatan individu dalam pembelajaran berkumpulan dalam kedua-dua sampel latihan dan ujian bagi semua kes set data regresi dan klasifikasi. Ia menunjukkan prestasi terbaik dalam sampel ujian bagi kedua-dua kumpulan set data regresi dan klasifikasi dengan masing-masing antara 0.86 hingga hampir 1 dan antara julat 0.89 hingga 1 bagi CA.

Effective fault detection and classification play essential roles in reducing the hazards such as electric shocks and fire in photovoltaic (PV) systems. However, the issues of interest in fault detection and classification for PV systems remain an open-ended challenge due to manual and time-consuming processes that require the relevant domain knowledge and experience of fault diagnoses. This paper proposes a hybrid deep-learning (DL) model-based combined architectures as the novel DL approaches to achieve the real-time automatic fault detection and classification of a PV system. This research employed the wavelet packet transform (WPT) as a data preprocessing technique to handle the PV voltage signal collected and feeding them as the inputs for combined DL architectures that consist of the equilibrium optimizer algorithm (EOA) and long short-term memory (LSTM-SAE) approaches. The combined DL architectures are able to extract the fault features automatically from the preprocessed data without requiring any previous knowledge, therefore can override the traditional shortages of manual feature extraction and manual selection of optimal features from the extracted fault features. These desirable features are anticipated to speed up the fault detection and classification capability of the proposed DL model with higher accuracy. In order to determine the performance of the proposed fault model, we carried out a comprehensive evaluation study on a 250-kW grid-connected PV system. In this paper, symmetrical and asymmetrical faults have been studied involving all the phases and ground faults such as single phase to ground, phases to phase, phase to phase to ground, and three-phase to ground. The simulation results validate the efficacy of the proposed model in terms of computation time, accuracy of fault detection, and noise robustness. Comprehensive comparisons between the simulation results and previous studies demonstrate the multidisciplinary applications of the present study.

Kelantan river (Sungai Kelantan in Malaysia) basin is one of the essential catchments as it has a history of flood events. Numerous studies have been conducted in river basin modelling for the prediction of flow and mitigation of flooding events as well as water resource management. Therefore, having multi-step-ahead forecasting for river flow (RF) is of important research interest in this regard. This study presents four different approaches for multi-step-ahead forecasting for the Kelantan RF, using NARX (nonlinear autoregressive with exogenous inputs) neural networks and deep learning recurrent neural networks called LSTM (long short-term memory). The dataset used was obtained in monthly record for 29 years between January 1988 and December 2016. The results show that two recursive methods using NARX and LSTM are able to do multi-step-ahead forecasting on 52 series of test datasets with NSE (Nash–Sutcliffe efficiency coefficient) values of 0.44 and 0.59 for NARX and LSTM, respectively. For few-step-ahead forecasting, LSTM with direct sequence-to-sequence produces promising results with a good NSE value of 0.75 (in case of two-step-ahead forecasting). However, it needs a larger data size to have better performance in longer-step-ahead forecasting. Compared with other studies, the data used in this study is much smaller. HIGHLIGHTS Four different approaches for multi-step-ahead forecasting for river flow.; Two forecasting approaches are performed in multivariate approach using NARXNN (nonlinear autoregressive with exogenous inputs - neural networks) with respectively recursive and direct auto-regressive mode.; Two forecasting approaches are performed using LSTM (long short-term memory) with respectively recursive univariate and sequence-to-sequence multivariate approach.; LSTM with a direct sequence-to-sequence multivariate approach performs the best for few-step-ahead forecasting in the limited data size. It has promising application for longer-step-ahead forecasting provided that the data size is sufficiently large.;

This paper addresses a design optimization of a gas turbine (GT) for marine applications. A gain-scheduling method incorporating a meta-heuristic optimization is proposed to optimize a thermodynamics-based model of a small GT engine. A comprehensive control system consisting of a proportional integral (PI) controller with additional proportional gains, gain scheduling, and a min-max controller is developed. The modeling of gains as a function of plant variables is presented. Meta-heuristic optimizations, namely a genetic algorithm (GA) and a whale optimization algorithm (WOA), are applied to optimize the designed control system. The results show that the WOA has better performance than that of the GA, where the WOA exhibits the minimum fitness value. Compared to the unoptimized gain, the time to reach the target of the power lever angle is significantly reduced. Optimal gain scheduling shows a stable response compared with a fixed gain, which can have oscillation effects as a controller responds. An effect of using bioethanol as a fuel has been observed. It shows that for the same input parameters of the GT dynamics model, the fuel flow increases significantly, as compared with diesel fuel, because of its low bioethanol heating value. Thus, a significant increase occurs only at the gain that depends on the fuel flow.

Water resources management in Malaysia has become a crucial issue of concern due to its role in the economic and social development of the country. Kelantan river (Sungai Kelantan) basin is one of the essential catchments as it has a history of flood events. Numerous studies have been conducted in river basin modelling for the prediction of flow and mitigation of flooding events as well as water resource management. This paper presents river flow modelling based on meteorological and weather data in the Sungai Kelantan region using a cascade-forward neural network trained with particle swarm optimization algorithm (CFNNPSO). The result is compared with those trained with the Levenberg–Marquardt (LM) and Bayesian Regularization (BR) algorithm. The outcome of this study indicates that there is a strong correlation between river flow and some meteorological and weather variables (weighted rainfall, average evaporation and temperatures). The correlation scores (R) obtained between the target variable (river flow) and the predictor variables were 0.739, −0.544, and −0.662 for weighted rainfall, evaporation, and temperature, respectively. Additionally, the developed nonlinear multivariable regression model using CFNNPSO produced acceptable prediction accuracy during model testing with the regression coefficient (R2), root mean square error (RMSE), and mean of percentage error (MPE) of 0.88, 191.1 cms and 0.09%, respectively. The reliable result and predictive performance of the model is useful for decision makers during water resource planning and river management. The constructed modelling procedure can be adopted for future applications.

The aim of this paper is to test the ability of artificial neural network (ANN) as an alternative method in time series forecasting and compared to autoregres­sive integrated moving average (ARIMA) in studying saving deposit in Malay­sian Islamic banks. Artificial neural network is getting popular as an alterna­tive method in time series forecasting for its capability to capture vola­tility pattern of non-linear time series data. In addition, the use of an estab­lished tool of analysis such as ARIMA is of importance here for comparative purposes. These two methods are applied to monthly data of the Malaysian Islamic bank­ing deposits from January 1994 to November 2005. The result provides evidence that ANN using “early stopping” approach can be used as an alterna­tive forecasting engine with univariate time series model. It can predict non-lin­ear time series using the pattern of the data directly without any statisti­cal analysis.

The aim of this paper is to use, compare, and analyze two forecasting technique: namely Auto Regressive Integrated Moving Average(ARIMA) and Artificial Neural Network(ANN) using Kuala Lumpur Composite Index(KLCI) in Malaysia. Artificial Neural Network is used because of its popularity of capturing the volatility patterns in nonlinear time series while ARIMA used since it is a standard method in the forecasting tool. Daily data of Kuala Lumpur Composite Index from 4 January 1999 to 26 September 2005 is used. ANN training with “early stopping” technique is investigated. We found that the deviation or error showed in the ANN technique is much less than that in ARIMA. Hence ANN can be used as a good forecaster engine for univariate time series model. It can predict nonlinear time series using the pattern of the past data. The proposed technique may help government, decision makers and planners especially in Malaysia. Keyword : Auto Regressive Integrated Moving Average(ARIMA), artificial neural network(ANN) and Kuala Lumpur Composite Index(KLCI)