International Research Journal of Education and Technology

Convolutional neural networks are being used by vision-based systems to detect fire during surveillance thanks to recent developments in embedded processing (CNNs). However, the application of such algorithms in surveillance networks is constrained by the fact that they typically need more processing time and memory. In this study, we suggest a CNN architecture for surveillance movies that can efficiently identify fire. Given its fair computing complexity and adaptability for the desired purpose in comparison to other computationally expensive networks like AlexNet, the GoogleNet architecture served as the model's inspiration. The model is adjusted taking into account the nature of the target problem and fire data in order to strike a compromise between efficiency and accuracy. The efficiency of the proposed framework is demonstrated by experimental findings on benchmark fire datasets, which also confirm its appropriateness for fire detection in CCTV surveillance systems.