Research On Cruise Fire Detection Technology Based On Multi-sensor Information Fusion

With China's goal of becoming one of the most dynamic markets in the global cruise market,cruise safety,especially fire safety,has also become the focus of attention.Unlike traditional fire detection,cruise ship fire detection needs to comprehensively process a large number of different types of data to achieve fire location detection and hazard classification of different types of cabins.At present,there are still some difficulties in how to properly use and fuse these data,mainly including: generalization of fire target detection,multi-camera information fusion and multi-sensor information fusion.Therefore,this article studies the above three problems.1.This article first applies the target detection algorithm to visible light image fire detection.First,it briefly introduces the currently used target detection algorithms,analyzes the Faster R-CNN target detection algorithm in detail,and then passes Obtain visible light video materials and process them into image frames and label categories to make data sets.Then the data set is sent to Faster R-CNN for training test.The test found that for visible light images,Faster R-CNN target detection algorithm can effectively distinguish the stages of flames and detect the positioning,but it cannot detect flames in occluded and smoldering flames,which has certain limitations.2.In order to solve the above-mentioned limitations,a thermal imaging image is introduced,but due to the shortcomings such as low contrast of the thermal imaging image,the direct fusion effect with the visible light image is very poor.First use the improved histogram equalization to enhance the thermal imaging image to improve its image contrast;then use the pseudo-colorization process to obtain a color image of the thermal imaging image;then,according to the camera imaging model,establish a thermal imaging map that maps to visible light Image transformation model;Finally,the fused image is put into Faster R-CNN for training and testing.The test verified that the visible light image combined with the thermal imaging image can effectively distinguish the different stages of the flame,even the smoldering and residual heat states,providing a basis for later assessment of the fire level of the cabin.3.The assessment of the fire danger level needs to integrate multiple types of information for comprehensive judgment.Because the cost of the fire spread experiment is too high,the data is obtained through simulation experiments in this paper.Firstly,qualitative and quantitative analysis was performed on the spread of smoke with Pyrosim software.Sixteen parameters were selected from the fire compartment and the remaining compartments as raw data;then 1 × 1 convolution,multi-convolution kernel and Channel weight allocation and other methods improve the fully connected neural network.The improved model has improved accuracy by 2.1% to 91.25%,but the parameters are only 1/10 of the fully connected network,which greatly reduces the storage requirements of the model and provides a certain basis for future embedded platform development.