Edge-AI Radiation Monitoring System: Deployment on Qualcomm Dual-Processor, Ultra-Low-Power ARM with TinyML, and FPGA Platforms

{"This":[0,114],"paper":[1],"presents":[2],"a":[3,17,66],"comprehensive":[4],"Edge-AI":[5],"framework":[6],"for":[7,32,60,71,141,218,238,264,273,285],"radiation":[8,34],"monitoring":[9,240],"built":[10],"around":[11],"three":[12],"complementary":[13],"embedded":[14],"platforms,":[15],"with":[16,65,138,228,301],"primary":[18],"focus":[19],"on":[20,296],"exploiting":[21],"the":[22,28,43,94,195],"real-time":[23,62,127,207,248],"microcontroller":[24,59],"(MPU)":[25],"capabilities":[26],"of":[27,116,202,250],"Arduino":[29,49],"UNO":[30,50],"Q":[31],"precise":[33],"sensor":[35,283],"control,":[36,86],"while":[37,125,152,193],"offloading":[38],"machine":[39,216],"learning":[40,217],"inference":[41,140],"to":[42,170],"GPU-accelerated":[44,67],"Qualcomm":[45,68,95,297],"AI":[46,69],"subsystem.":[47],"1.":[48],"Q:":[51],"A":[52,160,244],"heterogeneous":[53],"dual-processor":[54],"architecture":[55],"combining":[56],"an":[57,200],"STM32U585":[58],"deterministic":[61,123,172],"pulse":[63,82,209],"acquisition":[64,124],"subsystem":[70],"on-device":[72,215],"inference.":[73],"The":[74,175,204,260],"MPU":[75],"handles":[76],"all":[77],"time-critical":[78],"tasks,":[79],"including":[80,102,270],"radiation-sensor":[81],"detection,":[83,108,224],"SiPM":[84,212],"bias":[85],"temperature-compensated":[87],"high-voltage":[88],"regulation,":[89],"and":[90,109,143,165,186,221,232,241,257,267,281,303],"instrument":[91],"control.":[92],"Meanwhile,":[93],"GPU":[96],"executes":[97],"multiple":[98],"machine-learning":[99],"models":[100],"concurrently,":[101],"gamma-ray":[103],"spectral":[104,129],"classification,":[105],"cosmic-ray":[106,222],"track":[107,223],"camera-based":[110,225],"human":[111,226,255],"presence":[112],"detection.":[113],"division":[115],"labor":[117],"enables":[118,247],"true":[119],"concurrent":[120],"operation,":[121],"ensuring":[122],"delivering":[126],"AI-driven":[128],"intelligence.":[130],"2.":[131],"Nicla":[132],"Vision:":[133],"An":[134],"ultra-low-power":[135],"dual-ARM":[136],"system":[137,205,261],"TinyML":[139],"wearable":[142,271],"portable":[144],"dosimetry":[145],"applications,":[146,269],"consuming":[147],"less":[148],"than":[149],"150":[150],"mW":[151],"achieving":[153],"classification":[154,191,220,253],"accuracies":[155],"exceeding":[156],"95%.":[157],"3.":[158],"FPGA:":[159],"hardware-accelerated":[161],"implementation":[162],"using":[163],"hls4ml":[164],"Vitis,":[166],"leveraging":[167],"massive":[168],"parallelism":[169],"achieve":[171],"sub-microsecond":[173],"latency.":[174],"compressed":[176],"student":[177],"model":[178],"was":[179],"optimized":[180],"through":[181],"knowledge":[182],"distillation,":[183],"8-bit":[184],"quantization,":[185],"structured":[187],"pruning,":[188],"preserving":[189],"100%":[190],"accuracy":[192],"reducing":[194],"parameter":[196],"count":[197],"by":[198],"nearly":[199],"order":[201],"magnitude.":[203],"integrates":[206],"digital":[208],"processing,":[210],"closed-loop":[211],"temperature":[213],"compensation,":[214],"isotope":[219],"detection":[227],"distance-aware":[229],"dose-rate":[230,258],"estimation,":[231],"IoT":[233],"connectivity":[234],"(LoRaWAN,":[235],"WiFi,":[236],"MQTT)":[237],"remote":[239],"alert":[242],"generation.":[243],"web-based":[245],"interface":[246],"visualization":[249],"gamma":[251],"spectra,":[252],"results,":[254],"presence,":[256],"estimates.":[259],"is":[262],"designed":[263],"nuclear":[265,279],"safety":[266],"security":[268],"dosimeters":[272],"first":[274],"responders,":[275],"fixed":[276],"installations":[277],"at":[278],"facilities,":[280],"distributed":[282],"networks":[284],"border":[286],"monitoring.":[287],"Cite":[288],"as:M.":[289],"Altayeb,":[290],"\\"Edge-AI":[291],"Radiation":[292],"Monitoring":[293],"System:":[294],"Deployment":[295],"Dual-Processor,":[298],"Ultra-Low-Power":[299],"ARM":[300],"TinyML,":[302],"FPGA":[304],"Platforms,\\"":[305]}

Authors

Publication Details

Journal
Zenodo (CERN European Organization for Nuclear Research)
Published
2026-09-09
DOI
https://doi.org/10.5281/zenodo.22681374
Primary Topic
Radiation Detection and Scintillator Technologies
Type
preprint
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
preprint

Edge-AI Radiation Monitoring System: Deployment on Qualcomm Dual-Processor, Ultra-Low-Power ARM with TinyML, and FPGA Platforms

Moez Altayeb
Zenodo (CERN European Organization for Nuclear Research)
Radiation Detection and Scintillator Technologies
preprint

Edge-AI Radiation Monitoring System: Deployment on Qualcomm Dual-Processor, Ultra-Low-Power ARM with TinyML, and FPGA Platforms

Moez Altayeb
preprint en

Abstract

This paper presents a comprehensive Edge-AI framework for radiation monitoring built around three complementary embedded platforms, with a primary focus on exploiting the real-time microcontroller (MPU) capabilities of the Arduino UNO Q for precise radiation sensor control, while offloading machine learning inference to the GPU-accelerated Qualcomm AI subsystem. 1. Arduino UNO Q: A heterogeneous dual-processor architecture combining an STM32U585 microcontroller for deterministic real-time pulse acquisition with a GPU-accelerated Qualcomm AI subsystem for on-device inference. The MPU handles all time-critical tasks, including radiation-sensor pulse detection, SiPM bias control, temperature-compensated high-voltage regulation, and instrument control. Meanwhile, the Qualcomm GPU executes multiple machine-learning models concurrently, including gamma-ray spectral classification, cosmic-ray track detection, and camera-based human presence detection. This division of labor enables true concurrent operation, ensuring deterministic acquisition while delivering real-time AI-driven spectral intelligence. 2. Nicla Vision: An ultra-low-power dual-ARM system with TinyML inference for wearable and portable dosimetry applications, consuming less than 150 mW while achieving classification accuracies exceeding 95%. 3. FPGA: A hardware-accelerated implementation using hls4ml and Vitis, leveraging massive parallelism to achieve deterministic sub-microsecond latency. The compressed student model was optimized through knowledge distillation, 8-bit quantization, and structured pruning, preserving 100% classification accuracy while reducing the parameter count by nearly an order of magnitude. The system integrates real-time digital pulse processing, closed-loop SiPM temperature compensation, on-device machine learning for isotope classification and cosmic-ray track detection, camera-based human detection with distance-aware dose-rate estimation, and IoT connectivity (LoRaWAN, WiFi, MQTT) for remote monitoring and alert generation. A web-based interface enables real-time visualization of gamma spectra, classification results, human presence, and dose-rate estimates. The system is designed for nuclear safety and security applications, including wearable dosimeters for first responders, fixed installations at nuclear facilities, and distributed sensor networks for border monitoring. Cite as:M. Altayeb, "Edge-AI Radiation Monitoring System: Deployment on Qualcomm Dual-Processor, Ultra-Low-Power ARM with TinyML, and FPGA Platforms,"

Zenodo (CERN European Organization for Nuclear Research)
Radiation Detection and Scintillator Technologies
DOIOpen Access PDFOpenAlex
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.