Research And Application Of YOLOv5 Algorithm For Fire And Smoke Image Detection

With the conflagration occuring frequently,people’s life safety and propety has been facing huge threat.So the efficiency of detection and warning for fire plays a vital role in protecting people’s lives and property and keeping social harmony stability.The traditional sensor and image processing technology has some problems in practical application,such as high cost,high false alarm rate and small scope of action.In the current computer vision research field,using deep learning network to detect objects has become a hot research point.The YOLOv5 model has significant advantages in real-time detection compared to other deep learning object detection algorithms.In response to the low accuracy and easy loss of small targets caused by the application of the original model in fireworks detection,this thesis conducts research in the following aspects:1.In view of the fact that there is no ready-made public data set available for pyrotechnic detection,this thesis has collected and marked 19,800 pyrotechnic data sets from CSDN,Github,Baidu and other online platforms,which has laid a solid theoretical foundation for the development of pyrotechnic detection studied in this thesis.2.In order to improve the accuracy of flame and smoke detection,this thesis proposes an improved YOLOv5 algorithm for detection.Firstly,the cooperative attention block is loaded in the feature extraction part of YOLOv5.The cooperative attention block is equipped on the C3 layer of the feature extraction module to improve the sensitivity of the model to spatial information,reduce redundant information without losing feature information,and help the model locate the target more accurately.Secondly,the extra layers of small target detection is added to the feature fusion part,which include the relevant feature extraction and feature fusion modules.This improvement enables the model to obtain larger scale features,thus improving the ability to detect small targets.Finally,the boundary regression loss function S-IOU is equipped with this model to improve the detection accuracy.In the final network model,when the number of parameters increased by 9.3%,the m AP value increased by 3.4%,and the detection speed reached decreased little to realize the real-time detection of flame and smoke targets.3.In order to improve the usability,this thesis designed and implemented a set of fire and smoke detection system,using the open source Py Qt5 library to make a visual interface and using Python language to write backstage processing functions,easy to operate,and completed the function test of the system.