학술논문
CInet: Towards Precise Flame and Smoke Detection with Grid Partition Adopting Context and Interior Paths
Document Type
Conference
Author
Source
2022 IEEE Conference on Telecommunications, Optics and Computer Science (TOCS) Telecommunications, Optics and Computer Science (TOCS), 2022 IEEE Conference on. :1018-1023 Dec, 2022
Subject
Language
Abstract
Sensitive and timely flame and smoke detection in early fire warning is of great significance to environment and public security. The simultaneous detection of indoor and outdoor multi-scenario flame smoke has remained a challenging task due to the irregular properties of these objects. In this paper, we propose a precise fire smoke detection model-CInet(Context path and interior paths net), which is built upon two pathways. The Context path is designed to extract the global features of image. For the interior path, the grid partition technique is first applied to predict the class of each grid and to estimate the region of flaming smoke as a result, then interior path network is designed to extract each grid's interior features. Finally, the information from two pathways is integrated based on channel attention mechanism in order to characterize the feature of fire and smoke efficiently. Experimental results demonstrated encouraging performance when compared with other state-of-the-art methods in quantitative metrics. The proposed model is faster, and in the meantime, our model can significantly reduce the false alarm rate and detect the outline of flame and smoke in more detail.