Fire detection method and system based on deep learning, and storage medium

• Main Authors: DONG HONGLI, LU JINGYI, YANG DANDI, HU ZHONGRUI, ZHOU YINA, QIAN YING, WANG PENG, WANG DONGMEI
• Format: Patent
• Language: Chinese; English
• Published: 27.08.2024
• Subjects: ALARM SYSTEMS; CALCULATING; COMPUTER SYSTEMS BASED ON SPECIFIC COMPUTATIONAL MODELS; COMPUTING; COUNTING; ORDER TELEGRAPHS; PHYSICS; SIGNALLING; SIGNALLING OR CALLING SYSTEMS

The invention discloses a fire detection method and system based on deep learning, and a storage medium, relates to the technical field of target detection, and aims to solve the problems that feature extraction is difficult and the detection effect on small target flames is poor when target detection is carried out on variable fire in the prior art. A fire detection network model comprises a feature extraction backbone network, a multi-scale weighted feature fusion neck network and a target classification regression network. An MCA attention module with a plurality of different branches is introduced into the feature extraction backbone network to extract multi-level features, and a RepVB module is adopted in front-end convolution of the feature extraction backbone network to reduce model parameter quantity; features extracted in multiple stages in a backbone network are weighted and fused into a neck network through a multi-scale weighted feature fusion neck network, multiple features output by the neck net