IEEE Journal of Selected Areas in Sensors最新文献_第2页

FL-SQMPC: Secure Quantized Aggregation for Mitigating Gradient Inversion Attacks in Federated IoT Intrusion Detection FL-SQMPC：安全量化聚合缓解联邦物联网入侵检测中的梯度反转攻击

IEEE Journal of Selected Areas in Sensors Pub Date : 2026-01-01 Epub Date: 2026-08-18 DOI: 10.1109/JSAS.2026.3725190

Imene Bessaa;Elmahdi Bentafat;Zakaria Abou El Houda;Lyes Khoukhi

{"title":"FL-SQMPC: Secure Quantized Aggregation for Mitigating Gradient Inversion Attacks in Federated IoT Intrusion Detection","authors":"Imene Bessaa;Elmahdi Bentafat;Zakaria Abou El Houda;Lyes Khoukhi","doi":"10.1109/JSAS.2026.3725190","DOIUrl":"https://doi.org/10.1109/JSAS.2026.3725190","url":null,"abstract":"Federated Learning (FL) enables collaborative model training without centralizing raw data, but the exchange of model updates exposes clients to gradient inversion attacks (GIAs), which can reconstruct private training data from communicated gradients alone. To address this issue, we propose FL-SQMPC, a secure aggregation protocol that combines layer-wise mixed-precision quantization, fixed-point encoding, and additive secret sharing over a prime field. Under a noncollusion assumption on the secure aggregators, the secret-sharing layer provides information-theoretic hiding of each client update, while mixed-precision quantization reduces the precision and size of the transmitted update representation. We evaluate FL-SQMPC on Bot-IoT, ToN-IoT, and CICIoT2023 under Independent and Identically Distributed (IID) and non-IID partitions, comparing against vanilla FL, Differential Privacy, Homomorphic Encryption, and recent baselines, including CIDIoT. Under a TabLeak-based reconstruction adversary that observes the protected quantized update in the stronger colluding-participant setting, FL-SQMPC keeps tabular feature-recovery accuracy below vanilla FL across most evaluated settings. On predictive utility, FL-SQMPC matches vanilla FL within reproducibility noise and outperforms CIDIoT by up to <inline-formula><tex-math>$+5.54%$</tex-math></inline-formula> across the non-IID multiclass settings, while incurring substantially lower computational overhead than HE and CIDIoT. These results demonstrate a practical privacy–utility–efficiency tradeoff for resource-constrained IoT deployments.","PeriodicalId":100622,"journal":{"name":"IEEE Journal of Selected Areas in Sensors","volume":"3 ","pages":"338-352"},"PeriodicalIF":0.0,"publicationDate":"2026-01-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=11657663","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148986083","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Degraded Infrared Image Semantic Segmentation With Super-Resolution Tasks 基于超分辨率任务的退化红外图像语义分割

IEEE Journal of Selected Areas in Sensors Pub Date : 2026-01-01 Epub Date: 2026-07-01 DOI: 10.1109/JSAS.2026.3709202

Zhongwei Liu;Ziji Ma;Jianhua Huang;Zhiru Gu;Xun Shao

{"title":"Degraded Infrared Image Semantic Segmentation With Super-Resolution Tasks","authors":"Zhongwei Liu;Ziji Ma;Jianhua Huang;Zhiru Gu;Xun Shao","doi":"10.1109/JSAS.2026.3709202","DOIUrl":"https://doi.org/10.1109/JSAS.2026.3709202","url":null,"abstract":"Image segmentation methods based on artificial intelligence have been shown to facilitate visual scene understanding. However, these methods are difficult to handle degraded images with blurred edges and lost details, especially degraded infrared images of small targets. A multitask infrared small target segmentation method, Degradation Infrared Super Resolution for Semantic Segmentation, is proposed in this article. Super resolution can more accurately restore the edge details in the image through degradation representation, while Segment anything model performs feature segmentation by providing visual attention focus information. To obtain the maximum benefits of integration and fusion, the feature interaction mechanism of feature affinity (FA) module is introduced. FA module enables the intermediate layer features in the super-resolution reconstruction process to interact with the segmentation network by fully exploiting the edge information in the image. Experimental results demonstrate that our proposed semantic segmentation method achieves higher segmentation accuracy.","PeriodicalId":100622,"journal":{"name":"IEEE Journal of Selected Areas in Sensors","volume":"3 ","pages":"267-276"},"PeriodicalIF":0.0,"publicationDate":"2026-01-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=11592158","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148481291","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Video Data Regression Method for Physiological and Behavioral Cues Assessment Aimed to Contactless Mental Fatigue Classification 基于视频数据回归的生理和行为线索评估方法在非接触精神疲劳分类中的应用

IEEE Journal of Selected Areas in Sensors Pub Date : 2026-01-01 Epub Date: 2026-08-31 DOI: 10.1109/JSAS.2026.3728770

Batol Hamoud;Alexey Kashevnik

{"title":"Video Data Regression Method for Physiological and Behavioral Cues Assessment Aimed to Contactless Mental Fatigue Classification","authors":"Batol Hamoud;Alexey Kashevnik","doi":"10.1109/JSAS.2026.3728770","DOIUrl":"https://doi.org/10.1109/JSAS.2026.3728770","url":null,"abstract":"Mental fatigue of computer operators poses significant risks to productivity, safety, and well-being, yet existing monitoring solutions remain either intrusive (relying on wearable sensors) or subjective (based on self-reports). To address this challenge we propose video data regression method for physiological and behavioral cues assessment aimed to human mental fatigue classification. The method is fully contactless, it uses standard human facial video. The proposed method integrates video-based models for estimating blood pressure and oxygen saturation with temporal RGB signal dynamics and behavioral features describing eye, mouth, and head movements extracted using Google MediaPipe. To overcome the lack of objective ground truth, a previously validated sensor-trained model is employed to generate minute-by-minute fatigue labels that reflect the same underlying cognitive processes as wearable physiological measurements. Class imbalance is addressed through multiple balancing strategies, and several machine learning and deep learning models are systematically evaluated, with categorical boosting (CatBoost) using class weighting achieving the most robust performance. Feature importance analysis further identifies a compact yet highly informative feature subset, improving computational efficiency without sacrificing accuracy. Evaluated using leave-one-subject-out cross-validation on prolonged operator sessions, the proposed framework demonstrates robust subject-independent generalization, highlighting the potential of video-derived physiological regression combined with behavioral analysis as a scalable, nonintrusive solution for continuous mental fatigue assessment in real-world occupational environments.","PeriodicalId":100622,"journal":{"name":"IEEE Journal of Selected Areas in Sensors","volume":"3 ","pages":"379-394"},"PeriodicalIF":0.0,"publicationDate":"2026-01-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=11672791","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149133495","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Characterization and Multichannel Directional Contact Sensing on a Curved Surface Using Flexible Pressure-Sensor Arrays 基于柔性压力传感器阵列的曲面多通道定向接触传感特性研究

IEEE Journal of Selected Areas in Sensors Pub Date : 2026-01-01 Epub Date: 2026-09-08 DOI: 10.1109/JSAS.2026.3732275

Huang-Chia Shih;Min-Kie Cheng

{"title":"Characterization and Multichannel Directional Contact Sensing on a Curved Surface Using Flexible Pressure-Sensor Arrays","authors":"Huang-Chia Shih;Min-Kie Cheng","doi":"10.1109/JSAS.2026.3732275","DOIUrl":"https://doi.org/10.1109/JSAS.2026.3732275","url":null,"abstract":"This study presents an embedded flexible pressure-sensing baseball for multichannel finger–ball contact force-trend analysis. Two customized force-sensing resistor arrays were installed beneath the leather surface to monitor ten contact regions associated with the index and middle fingers. A compact analog front-end converts sensor impedance variations into 0–3 V signals, and an STM32-based acquisition unit produces synchronized 10-channel outputs at 180 Hz. The assessment was conducted in two stages. Stage 1 characterized the sensor and measurement chain through loading-jig verification, 19-point quasi-static calibration over 0.1–1.9 kgf, analog-front-end testing, and channel-wise calibration. Stage 2 evaluated the complete baseball under controlled ball-surface loading to determine whether distinct directional and spatial contact patterns could be reconstructed. The loading jig exhibited a transmission error of less than ±1%, and the analog-front-end outputs differed from the calculated values by less than ±1%. Controlled loading produced distinguishable channel-activation, directional force-redistribution, and sequential contact-transfer patterns across the two finger arrays. The 180-Hz output rate corresponds to a 5.56-ms interval and is suitable for synchronized trend observation; however, it is not claimed to reproduce the complete waveform or absolute peak force of the final 10–20-ms pitching-release event. The proposed system is therefore positioned as an embedded platform for sensor assessment and relative contact-force trend analysis. Untethered integration, dynamic reference calibration, and human pitching trials remain future work.","PeriodicalId":100622,"journal":{"name":"IEEE Journal of Selected Areas in Sensors","volume":"3 ","pages":"468-483"},"PeriodicalIF":0.0,"publicationDate":"2026-01-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=11683660","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149133452","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Edge Imaging With a Low-Area, Low-Power 40 MS/s 0.001 mm$^{2}$ 5-Bit SAR 边缘成像与低面积，低功耗40 MS/s 0.001 mm$^{2}$ 5位SAR

IEEE Journal of Selected Areas in Sensors Pub Date : 2026-01-01 Epub Date: 2026-03-30 DOI: 10.1109/JSAS.2026.3678515

Sanchari Das;Soumyajit Mandal;Bhasker Choubey;Bibhu Datta Sahoo

{"title":"Edge Imaging With a Low-Area, Low-Power 40 MS/s 0.001 mm$^{2}$ 5-Bit SAR","authors":"Sanchari Das;Soumyajit Mandal;Bhasker Choubey;Bibhu Datta Sahoo","doi":"10.1109/JSAS.2026.3678515","DOIUrl":"https://doi.org/10.1109/JSAS.2026.3678515","url":null,"abstract":"Image sensors designed for edge applications are constrained by power requirements. Furthermore, a number of machine learning tasks can be implemented even with lower bit depth of images, thereby further reducing power requirements in conversion and transmission. Hence, classical 8 or more bit of column parallel analogue-to-digital converters (ADCs) may not be necessarily required in edge applications. In this work, we showcase this potential by designing a 5-bit successive approximation register (SAR) ADC with a sample rate of 40 MS/s while consuming of 380 <inline-formula><tex-math>$mu$</tex-math></inline-formula>W, an active area of 0.001 mm<inline-formula><tex-math>$^{2}$</tex-math></inline-formula> designed in a 65 nm complementary metal–oxide–semiconductor technology. The ADC achieves a signal-to-noise and distortion ratio of 25 dB, a peak signal-to-noise ratio (PSNR) of 32 dB, and a structural similarity index measure of 0.96 when digitizing the canadian institute for advanced research - 10 classes (CIFAR10) dataset. The ADC is also used for image classification tasks on modified national institute of standards and technology (MNIST), Fashion MNIST, extended modified national institute of standards and technology (EMNIST), CIFAR10, ImageNet, DIV2K, and labeled faces in the wild datasets. The accuracy of the digitized images closely matches the accuracy of the original images for all the datsets. The PSNR and the accuracy values show the effectiveness of the proposed ADC architecture for low-power and low-area machine vision applications.","PeriodicalId":100622,"journal":{"name":"IEEE Journal of Selected Areas in Sensors","volume":"3 ","pages":"171-179"},"PeriodicalIF":0.0,"publicationDate":"2026-01-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=11457606","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"147737112","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

TrustEvo: A Hierarchical Reinforcement Learning Task Offloading Framework With Trust Evolution for Social Internet of Things TrustEvo：基于社会物联网信任进化的分层强化学习任务卸载框架

IEEE Journal of Selected Areas in Sensors Pub Date : 2026-01-01 Epub Date: 2026-09-01 DOI: 10.1109/JSAS.2026.3728775

Boran Yang;Wei Zou;Jingjing Liu;Binqiang Ye;Dapeng Wu

{"title":"TrustEvo: A Hierarchical Reinforcement Learning Task Offloading Framework With Trust Evolution for Social Internet of Things","authors":"Boran Yang;Wei Zou;Jingjing Liu;Binqiang Ye;Dapeng Wu","doi":"10.1109/JSAS.2026.3728775","DOIUrl":"https://doi.org/10.1109/JSAS.2026.3728775","url":null,"abstract":"This article proposes TrustEvo, a novel hierarchical reinforcement learning framework for task offloading in the social Internet of Things (SIoT), aiming to achieve an effective balance between efficiency and trust in ubiquitous computing environments. SIoT is considered a key paradigm to enable decentralized and socially aware ubiquitous computing, where devices act as both task consumers and service providers. To address the challenges of dynamic environments, heterogeneous resources, and hybrid decision spaces, TrustEvo formulates task offloading as a hierarchical decision-making problem. Specifically, TrustEvo decomposes the offloading process into three coupled levels, jointly optimizing the offloading type, target node selection, and offloading ratio. This design enables efficient exploration of both discrete and continuous decision spaces while capturing the intrinsic structure of offloading decisions. TrustEvo explicitly integrates trust evolution into the decision-making loop. By modeling trust as a dynamic state variable driven by task execution outcomes, TrustEvo establishes a closed-loop interaction between system performance and node reliability, adaptively balancing the multiobjective optimization of latency minimization, success rate maximization, and trust value maximization under uncertain dynamic environments. Extensive experiments based on real-world datasets demonstrate that TrustEvo consistently outperforms baseline approaches. TrustEvo achieves lower and more stable latency, higher average trust levels, and robust task success rates. These results validate its effectiveness as a lightweight, adaptive, and trust-aware solution for next-generation SIoT-enabled ubiquitous computing systems.","PeriodicalId":100622,"journal":{"name":"IEEE Journal of Selected Areas in Sensors","volume":"3 ","pages":"395-410"},"PeriodicalIF":0.0,"publicationDate":"2026-01-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=11672792","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149130789","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Distributed Blind Equalization in Sensor Network via a Block-Combination Weight Strategy 基于块组合权策略的传感器网络分布式盲均衡

IEEE Journal of Selected Areas in Sensors Pub Date : 2026-01-01 Epub Date: 2026-06-08 DOI: 10.1109/JSAS.2026.3701622

Sulin Chi;Tetsuya Shimamura

{"title":"Distributed Blind Equalization in Sensor Network via a Block-Combination Weight Strategy","authors":"Sulin Chi;Tetsuya Shimamura","doi":"10.1109/JSAS.2026.3701622","DOIUrl":"https://doi.org/10.1109/JSAS.2026.3701622","url":null,"abstract":"Distributed estimation in wireless sensor networks has attracted considerable attention across various scientific and engineering disciplines. Traditionally, most studies have focused on estimating the unknown parameter of interest in noisy environments under the assumption that the training signal or the desired signal is known in advance at the receiver. However, in many practical applications, it is often difficult or even impossible to obtain the prior information of the training signal or the desired signal. Consequently, <italic>distributed blind equalization</i>, which seeks to recover the transmitted data signal without requiring prior knowledge of the training signal, has emerged as an important research topic. Nevertheless, the performance of the distributed blind equalization is highly sensitive to transmission channel conditions, particularly under ill-conditioned channels where severe noise or distortion may occur. In order to address this challenge, we propose a <italic>block-combination weight strategy</i> designed to mitigate the adverse effects of the ill-conditioned channel, thereby improving the accuracy of source signal estimation. A theoretical analysis of the proposed method is presented in terms of the mean-square error (MSE). Furthermore, the effectiveness of the proposed method is validated through computer simulations and compared with conventional methods using both MSE and symbol error rate (SER) as performance metrics. The simulation results demonstrate that the proposed approach consistently outperforms existing approaches in terms of both MSE and SER.","PeriodicalId":100622,"journal":{"name":"IEEE Journal of Selected Areas in Sensors","volume":"3 ","pages":"253-266"},"PeriodicalIF":0.0,"publicationDate":"2026-01-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=11554374","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148479141","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Large Language Models for Sensor Placement in Smart Spaces 智能空间中传感器放置的大型语言模型

IEEE Journal of Selected Areas in Sensors Pub Date : 2026-01-01 Epub Date: 2026-08-31 DOI: 10.1109/JSAS.2026.3729163

Yangyang Wang;Naser Hossein Motlagh;Agustin Zuniga;Huber Flores;Petteri Nurmi;Sasu Tarkoma

{"title":"Large Language Models for Sensor Placement in Smart Spaces","authors":"Yangyang Wang;Naser Hossein Motlagh;Agustin Zuniga;Huber Flores;Petteri Nurmi;Sasu Tarkoma","doi":"10.1109/JSAS.2026.3729163","DOIUrl":"https://doi.org/10.1109/JSAS.2026.3729163","url":null,"abstract":"Optimal sensor placement is vital for smart space monitoring, requiring a balance between data fidelity and signal stability. This study introduces an embodied layout reasoning framework, evaluating the zero-shot deployment capabilities of six state-of-the-art large language models (LMs) (e.g., ChatGPT-4o, Claude-Sonnet-4.5) acting as “virtual field engineers” within a university office. To quantify spatial hallucination and decision-making risks, each model generated placements 100 times. While Claude-Sonnet-4.5 demonstrated the highest intramodel consistency (59.7%), Gemini’s 9.9% reliability highlighted significant challenges in autonomous spatial grounding. Empirically, LLM-driven strategies achieved numerically stronger signal stability (mean received signal strength indicator: <inline-formula><tex-math>$-$</tex-math></inline-formula>49.04 dBm) compared with expert-designed strategies (<inline-formula><tex-math>$-$</tex-math></inline-formula>52.01 dBm), although this difference did not reach statistical significance. Furthermore, we establish a cross-modality spatial reasoning robustness framework, revealing that models like ChatGPT-4o and Grok can reconstruct geometric topologies from abstracted text or floor plans, enabling high-performance, privacy-preserving deployments. In addition, Doubao offers a favorable tradeoff between reasoning performance and operational cost. Finally, we propose a human–AI hybrid Intelligence paradigm; by transitioning from unconstrained inference to a two-stage workflow using a vetted candidate set, model consistency improved about 69.6%. Overall, our results suggest that combining LLM-based geometric exploration with human feasibility constraints yields robust, deployable sensor placement solutions for real-world Internet of Things environments.","PeriodicalId":100622,"journal":{"name":"IEEE Journal of Selected Areas in Sensors","volume":"3 ","pages":"439-454"},"PeriodicalIF":0.0,"publicationDate":"2026-01-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=11673158","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149133794","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

A Comparative Study of CNN Optimization Methods for Edge AI: Exploring the Role of Early Exits 边缘AI中CNN优化方法的比较研究：探索早期退出的作用

IEEE Journal of Selected Areas in Sensors Pub Date : 2026-01-01 Epub Date: 2026-09-07 DOI: 10.1109/JSAS.2026.3731343

Nekane Fernandez;Ivan Valdes;Julen Arratibel;Steven Van Vaerenbergh;Idoia de la Iglesia

{"title":"A Comparative Study of CNN Optimization Methods for Edge AI: Exploring the Role of Early Exits","authors":"Nekane Fernandez;Ivan Valdes;Julen Arratibel;Steven Van Vaerenbergh;Idoia de la Iglesia","doi":"10.1109/JSAS.2026.3731343","DOIUrl":"https://doi.org/10.1109/JSAS.2026.3731343","url":null,"abstract":"Deploying deep neural networks on edge devices requires balancing accuracy, latency, energy, and resource constraints under realistic execution conditions. To fit models within these constraints, two broad strategies have emerged: static compression techniques, such as pruning and quantization, which permanently reduce model size; and dynamic approaches, such as early-exit mechanisms, which adapt computational cost at runtime. While both families are widely studied in isolation, they are rarely compared under identical conditions on physical hardware. This article presents a unified, deployment-oriented comparison of static compression and dynamic early-exit mechanisms, evaluated on real edge devices using open neural network exchange (ONNX)-based inference pipelines. Our results show that static and dynamic compression techniques offer fundamentally different tradeoffs for edge deployment. While pruning and quantization deliver consistent memory footprint reduction, early-exit mechanisms enable input-adaptive computation savings that they cannot match. Their combination, however, proves particularly effective—reducing inference latency by up to 3.12× while retaining a 3.9× compression ratio with minimal accuracy loss, and providing meaningful energy savings in latency-critical configurations.","PeriodicalId":100622,"journal":{"name":"IEEE Journal of Selected Areas in Sensors","volume":"3 ","pages":"484-503"},"PeriodicalIF":0.0,"publicationDate":"2026-01-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=11680528","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149188544","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

IP-UNet: Intensity Projection UNet Architecture for IoMT-Enabled 3-D Medical Volume Segmentation IP-UNet：用于支持iomt的三维医疗体分割的强度投影UNet体系结构

IEEE Journal of Selected Areas in Sensors Pub Date : 2026-01-01 Epub Date: 2026-03-31 DOI: 10.1109/JSAS.2026.3679493

Nyothiri Aung;Mohand Tahar Kechadi;Liming Chen;Sahraoui Dhelim

{"title":"IP-UNet: Intensity Projection UNet Architecture for IoMT-Enabled 3-D Medical Volume Segmentation","authors":"Nyothiri Aung;Mohand Tahar Kechadi;Liming Chen;Sahraoui Dhelim","doi":"10.1109/JSAS.2026.3679493","DOIUrl":"https://doi.org/10.1109/JSAS.2026.3679493","url":null,"abstract":"The proliferation of high-resolution medical imaging data, increasingly generated and transmitted via Internet of Medical Things (IoMT) infrastructure, presents significant computational challenges. Limited memory capacity remains a common drawback when processing these high-resolution 3-D volumetric datasets, as direct cropping or down-sizing often results in a loss of critical diagnostic detail, exacerbates class imbalance, and degrades segmentation performance. 3-D volumes are usually cropped or downsized first before processing, which can result in a loss of resolution, increase class imbalance, and affect the performance of the segmentation algorithms. In this article, we propose an end-to-end deep learning approach called IP-UNet. IP-UNet is a UNet-based model that performs multiclass segmentation on intensity projection (IP) of 3-D volumetric data instead of the memory-consuming 3-D volumes. IP-UNet uses limited memory capability for training without losing the original 3-D image resolution. We compare the performance of three models in terms of segmentation accuracy and computational cost. 1) Slice-by-slice 2-D segmentation of the CT scan images using a conventional 2-D UNet model. 2) IP-UNet that operates on data obtained by merging the extracted maximum intensity projection (MIP), closest vessel projection (CVP), and average intensity projection (AvgIP) representations of the source 3-D volumes, then applying the UNet model on the output IP images. 3) 3-D-UNet model directly reads the 3-D volumes constructed from a series of CT scan images and outputs the 3-D volume of the predicted segmentation. We test the performance of these methods on 3-D volumetric images. Experimental results show that IP-UNet can achieve similar segmentation accuracy with 3-D-UNet but with much better performance. It reduces the training time by 70% and memory consumption by 92%.","PeriodicalId":100622,"journal":{"name":"IEEE Journal of Selected Areas in Sensors","volume":"3 ","pages":"191-198"},"PeriodicalIF":0.0,"publicationDate":"2026-01-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=11458901","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"147796176","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0