Research On Disaster Image Recognition And Location Technology

Fire is a kind of disaster that often occurs and seriously threatens public safety and social development.Smoke usually precedes flames in the early stages of a fire,so research on smoke detection is of great significance.Smoke detectors that are detected based on the smoke and temperature changes that accompany a fire generally can only alarm when the smoke concentration reaches a corresponding value.It is not suitable for smoke detection in large spaces and has poor real-time performance.Compared with traditional fire detectors,fire detection methods that use image recognition technology to detect fire conditions and use target tracking technology to locate specific locations are more real-time and reliable.Therefore,the research on smoke image recognition and target tracking technology has important value.Two smoke image detection and an improved target tracking technology are proposed in this thesis.By simulation experiments in a public database,the results show that the method proposed in this thesis can achieve better results.The main work and innovations of this thesis are as follows:Firstly,for the smoke detection problem,a smoke detection method based on improved multi-feature fusion is proposed.Because smoke shapes,colors,and textures vary widely,it is difficult to accurately and comprehensively describe the characteristics of smoke by extracting a single feature.Therefore,a method for extracting the HOG features,LBP features of smoke images and their depth features with the VGG16 network.By combining these three features,implementing smoke detection with a fully connected neural network classification algorithm is proposed.Experimental results show that the detection effect of this method has been significantly improved.Secondly,a video smoke detection method based on an improved background model is proposed.The traditional smoke detection method used the ambiguity of smoke for discrimination.Because the background extraction effect is not good,it affects the accuracy of fire smoke detection.The method in this thesis first uses the idea that the gray value of similar objects does not change much,divides different statistical modules according to the gray histogram information of the video image,and merges them into a background image.Then,the two-dimensional wavelet transform method is used to extract the blurring characteristics of the smoke background.The energy value of the background composite image is compared with the energy value of the composite image when the smoke appears,and a threshold is set to detect the presence of smoke.The simulation results show that the improved method has higher detection accuracy.Thirdly,for the problem of target tracking,a particle filter tracking method based on improved multi-feature fusion is proposed.The color feature represents the overall description of the target object,while the HOG feature contains certain structural information,and complementing the two features can describe the feature more comprehensively.Aiming at the problem that the tracking algorithm based on the single-feature-based particle filter will lead to poor tracking results in complex environments,by merging the color and HOG features,a new fusion feature is formed for object tracking.Experimental results show that the improved target tracking method in this thesis can achieve good tracking results.