A Review of various Deep Learning Models for Fire Detection

Abstract

In recent years, the deep learning (DL) computing paradigm has emerged as the gold standard within the machine learning (ML) community. Gradually, it has become the predominant ML computational approach, yielding remarkable results across various intricate cognitive tasks, often rivalling or surpassing human performance. One of the paramount advantages of DL is its capacity to glean insights from massive datasets. Machine learning tools, constituting algorithmic applications of artificial intelligence, endow systems with the capability to learn and enhance themselves with minimal human intervention. Concepts such as data mining and predictive modelling align closely with this paradigm, facilitating software to refine its predictive accuracy without explicit programming. This paper introduces convolutional neural networks (CNNs), the most prevalent DL network type, and delineates the evolution of CNN architectures alongside their salient features. The progression is elucidated by commencing with seminal works like the AlexNet network and culminating with advanced architectures such as the High-Resolution network (HR.Net).

Furthermore, the paper discusses prevalent challenges encountered in DL research and proposes potential solutions to bridge existing gaps in understanding. Subsequently, a compendium of major DL applications is presented. The computational tools underpinning DL, including FPGA, GPU, and CPU, are summarized, underscoring their pivotal role in advancing DL methodologies