Object detection plays a crucial role in image processing and data analysis by enabling the extraction of visual information that supports decision-making across diverse practical domains. Applications of object detection span intelligent healthcare, autonomous vehicles, surveillance, robotic vision, and precision agriculture. Acceleration of object detection using convolutional neural networks can be examined from three complementary perspectives: structural, algorithmic, and hardware, which respectively reduce model complexity, optimize computational kernels, and exploit specialized processors. Numerous studies have proposed hardware accelerator architectures to improve the computational efficiency of You Only Look Once object detection models. This paper presents a systematic literature review of hardware accelerators for sparse and compressed YOLO models. The review analyzes how compression techniques, sparsity patterns, convolution implementation strategies, and hardware dataflows interact in YOLO acceleration. It also organizes commonly used evaluation metrics into five groups: sparsity and skipped-data ratio, effective computation reduction, memory and compression efficiency, hardware utilization and workload balance, and system-level efficiency with accuracy trade-off. The findings highlight that sparse data alone does not guarantee acceleration; practical efficiency depends on the co-design of sparsity pattern, convolution mapping, memory hierarchy, and processing-element scheduling. Finally, this review identifies research opportunities in structured sparsity, sparse data movement, processing-element utilization, load balancing, memory aware scheduling, and hardware software co-design for recent YOLO variants.
D. Endrawati, I. Syafalni, Nana Sutisna et al.· IEEE Access· 0 citations
Respiratory rate (RR) is a vital physiological marker whose accurate and continuous monitoring holds particular clinical significance. Conventional contact-based sensors, while reliable, impose physical constraints that may cause discomfort and skin irritation. Radar-based sensing, particularly frequency-modulated continuous-wave (FMCW) radar, offers a compelling contactless alternative: it operates through clothing, remains unaffected by lighting conditions, and enables unobtrusive continuous monitoring without compromising patient comfort or privacy. Nevertheless, radar-based RR estimation remains technically challenging due to signal noise, motion artifacts, and the inherent subtlety and variability of breathing patterns. Although prior work has demonstrated viable RR estimation using either signal processing or machine learning approaches, a direct comparison of both paradigms on pediatric radar data under a unified preprocessing framework has not been established. This study addresses that gap through a systematic and comprehensive comparative evaluation of classical signal processing and machine learning (ML) techniques for RR estimation using a publicly available FMCW radar dataset collected from 50 children under 13 years old. Six methods were evaluated: fast Fourier transform (FFT)-based frequency-domain analysis, time-domain peak detection, convolutional neural networks (CNN), multilayer perceptron (MLP), LightGBM regression, and KalmanNet. All were benchmarked against the ground truth obtained from a clinically validated reference system. The frequency-domain analysis method combined with multichannel averaging achieved the best overall performance, with a mean absolute error (MAE) of 2.60 breaths per minute (bpm), a root mean square error (RMSE) of 3.39 bpm, and an average inference time of 44.2 ms per analysis window, outperforming time-domain peak detection (MAE 3.66 bpm), CNN (MAE 2.75 bpm, RMSE 3.39 bpm), MLP (MAE 4.01 bpm, RMSE 4.64 bpm), LightGBM (MAE 5.91 bpm, RMSE 6.41 bpm), and KalmanNet (MAE 5.97 bpm, RMSE 6.73 bpm). These results indicate that, in the limited-data pediatric setting, a carefully designed classical signal processing method provides highly competitive performance, achieving accuracy statistically comparable to CNN while outperforming the remaining handcrafted feature-based learning methods, and remaining computationally efficient and interpretable.
Nur Ahmadi, H. V. Tanoto, Diyah Widiyasari et al.· IEEE Access· 0 citations
Myocardial infarction (MI) is a leading global health concern, typically diagnosed using ECG, biomarkers, or imaging, which costly or unavailable in low-resource settings. This study presents a non-invasive, machine learning-based approach using phonocardiogram (PCG) signals for classifying normal, ST-elevation MI (STEMI), and non-ST-elevation MI (NSTEMI). The proposed approach leverages the acoustic signatures of cardiac mechanical activity, capturing subtle variations in heart sound morphology and timing associated with ischemic myocardial dysfunction. The processing pipeline included PCG acquisition via electronic stethoscope, band-pass filtering, envelope-based segmentation, feature extraction, and selection using Mutual Information and K-best ranking. We evaluated the method using a diverse dataset of 104 subjects from Indonesia and Japan to ensure generalizability across ethnic and physiological variations. Eighteen key features with mutual information values up to 0.82 bits were used to train AdaBoost and Gradient Boosting models. Without parameter tuning, these models achieved 88.30% and 93.00% accuracy, respectively. After optimization, AdaBoost reached 94.00% accuracy, and Gradient Boosting achieved 98.30% accuracy and a 95.10% F1-score. These results outperform previous bagging-based methods of 86.00% accuracy, demonstrating improved accuracy and robustness. This work highlights the potential of PCG-based MI detection as a low-cost, non-invasive diagnostic alternative, particularly valuable for early screening and triage in resource-limited healthcare settings.
Ira Puspasari, Nobuo Watanabe, Masahiro Ohwada et al.· Scientific Reports· 0 citations
We use cookies to run the site and, with your consent, for analytics and to show ads.
See our Cookie Policy.
