Research And Implementation Of Fire Detection Based On Deep Learning

Fire detection is very important in safety inspection,people can effectively reduce the loss based on early warning.Traditional fire detection methods have obvious limitations.First of all,people need to use temperature detectors,smoke detectors,thermal cameras and other equipment to calculate the indoor temperature,the density of micro particles and smoke to check the fire,but these equipment are usually expensive,vulnerable,low accuracy and accompanied by too many false positives.Secondly,the traditional detection methods require developers to master the relevant professional knowledge such as the temperature,movement form and composition of flame and smoke,but these requirements seriously restrict the establishment of fire detection model.The advent of deep learning provides a new opportunity for fire detection.People only need to provide a large number of labeled data sets to train the model to automatically identify the flame and smoke from the data,while developers no longer need professional instruments and complete professional background knowledge.This paper mainly selects four commonly used object detection algorithms based on deep learning,namely:fast RCNN,r-fcn,SSD and Yolo.This paper uses the same data set to compare the performance of the four methods,and the results show that the Yolo algorithm is better than the other three algorithms in accuracy and detection time.Therefore,this paper further improves and experiments on the basis of the Yolo algorithm.This paper is mainly based on the latest yolov5 algorithm for experiments,and aiming at the two problems of incomplete coverage of the target border and poor detection effect in complex background,the overall accuracy of yolov5 algorithm is further improved.In this paper,we change the data set,add ECA(efficient channel attention)to the residual module,and use focal loss as the loss function.The experimental results show that the above improvement effectively increases the robustness of the model and significantly improves the detection accuracy.Finally,the experiment uses pyqt5 to make the fire detection system,encapsulates the improved project,and the detection results appear after the user uploads the pictures to be inspected.The system can upload pictures,detect targets and save results.