Λεπτομέρειες βιβλιογραφικής εγγραφής
| Τίτλος: |
An FPGA-Based Hardware Architecture for Automated Fire Detection. |
| Συγγραφείς: |
Ngo, Hieu-Truong1 truongnh@uit.edu.vn, Nguyen, Xuan-Dung2 23520339@gm.uit.edu.vn |
| Πηγή: |
IAENG International Journal of Computer Science. Jun2026, Vol. 53 Issue 6, p2226-2234. 9p. |
| Θεματικοί όροι: |
Field programmable gate arrays, Fire detectors, Convolutional neural networks, Computer architecture, OpenCL (Computer program language), Real-time computing |
| Περίληψη: |
Fire incidents pose serious risks, leading to substantial economic losses and threats to human life. Conventional fire-alarm systems based on physical sensors are widely deployed, yet they may miss small flames at an early stage, particularly in complex environments. This paper presents a field-programmable gate array (FPGA)-based hardware architecture for automated fire detection using a streamlined AlexNet convolutional neural network (CNN). The system is implemented on an Intel Arria 10 System on Chip (SoC) FPGA using the Open Computing Language (OpenCL)-based PipeCNN framework, an OpenCL-based CNN accelerator framework. To improve real-time performance, we introduce a dual-system pipelined execution scheme and customized memory-access patterns. Compared with the baseline configuration, the proposed design reduces the number of model parameters from 57 million to 25 million and shortens the end-to-end processing time from 155 milliseconds to 115 milliseconds. Experimental results show a detection accuracy of 86.88 percent on a balanced 5000-image test set and an average accuracy of 93 percent on a 27-video test set, while sustaining 15 frames per second and using only 8 percent of the FPGA digital signal processing (DSP) resources. These results indicate that the proposed architecture is suitable for real-time fire surveillance under resource and power constraints. [ABSTRACT FROM AUTHOR] |
| Βάση Δεδομένων: |
Supplemental Index |