Research And Implementation Of Smoke And Fire Detection Algorithm Based On Deep Learning

Fires are prone and severely destructive and pose a major threat to human property and life.It is of great significance to find the fire point and alarm timely and accurately for maintaining the normal production and life order.The traditional smoke and fire detection algorithms have the problems of complex design,slow calculation speed,easy to produce false positives and false negatives.In recent years,the deep learning methods have been applied to the field of computer vision,and achieved remarkable results.In order to improve the effect of smoke and fire detection in video,deep learning techniques are applied to the detection of smoke and fire,and the specific research contents are as follows:Firstly,the algorithm of smoke and fire detection based on SSD is studied.The SSD model adopts the idea of predicting multi-scale features and can detect objects of multiple size.The paper applies the SSD algorithm to the task of smoke and fire detection,constructs a smoke and fire dataset based on real scenes and uses color model conversion and the histogram equalization algorithm to normalize the illuminance of the images in the data set to standardize the characteristics of dataset.It improves the training efficiency of the model.Experimental results show that the SSD algorithm has a good detection effect on smoke and fire objects,but there are some shortcomings.Subsequently,a smoke and fire detection algorithm based on improved SSD is proposed to solve the problem that small-size smoke and fire objects detection is not effective and can easily produce false positives and false negatives.DenseNet is used as the basic network of SSD to improve its ability to detect small objects.Secondly,the focal loss is introduced into the loss function to solve the problem of positive and negative examples imbalance.It improves the robustness of the algorithm by increasing the weight of the hard examples in the loss function.The improved SSD model is trained by the smoke and fire dataset.Experiments indicate that this algorithm improves the detection effect of small-size smoke and fire objects considering the detection speed compared with current major detection algorithms.Finally,in order to achieve better smoke and fire detection effect in video,a video-based smoke and fire detection system is designed and implemented.The system consists of a video image processing module,a single frame smoke and fire detection module and a video frame correction module.The system was tested by using multiple scene videos.Experiments show that the system has good smoke and fire detection effect and can meet the requirements of industrial applications.