Based On The Improved YOLOv5 Smoke Detection System Research And Implementation

With the continuous updating and iteration of related technologies in the field of computer vision,although the current video-based smoke detection technology has made some progress,there are still problems such as insufficient detection accuracy and low detection rate.First of all,the currently public fire smoke datasets scenario is very single.In view of the poor performance of the fire smoke detection model trained by a single data set in terms of scene adaptation,and often unable to match some scenarios,poor universality,etc.,this study collects and organizes fire smoke datasets from different sources,including forests,factories,buildings,and vehicle fires.Aiming at the problems of insufficient accuracy and low detection rate of the existing fire smoke detection model,an improved fire smoke detection algorithm model based on YOLOv5 is proposed.The model adds CBAM attention mechanism to the network of the original YOLOv5 model to improve the accuracy of model detection.Replace the original activation function of the model with the ACON activation function,which increases the depth of the model network by setting selective activation neurons,improves the output ability and learning effect of the model,and reduces the false detection rate of the model;The EIo U loss function is used instead of the original loss function to improve the convergence speed and detection accuracy of the model.The average accuracy rate of the algorithm on the test set reached 96.5%,the accuracy rate and recall rate reached 95.1% and 94.8%,respectively,the average accuracy rate on the verification set reached 94.9%,and the accuracy rate and recall rate reached 94.2%and 93.3%,respectively.Finally,relying on the algorithm model of this study,a fire smoke detection system was developed,which realized the real-time fire smoke detection function,and in the system test,a 10 s outdoor fire video was detected,and it took 0.8s to successfully detect the location of the image fire smoke,achieving a relatively good fire smoke detection effect.