Research On Fire Smoke Detection Algorithm Of Ship Cabin Based On Video

Once a ship fires,it will directly threaten the lives of the people on board,the safety of the ship and the cargo carried,and may even cause inestimable environmental damage.Therefore,it is necessary to study a feasible and efficient ship fire detection system to ensure the safety of life and property of ship operators.Due to the special environment of ships,traditional fire detection systems based on sensors are prone to omissions and false alarms.In recent years,some researchers have combined computer image processing technology and deep learning technology with fire detection,that is,video-based fire detection algorithms,which have greatly promoted the development of fire detection systems.The video-based fire detection system has the advantages of large monitoring range,high degree of intelligence,not easily affected by the surrounding environment,and easy post-event video query,but there is still much room for improvement in the accuracy of fire detection.This paper introduces the video-based fire detection algorithm into the ship cabin fire detection system,and proposes a video-based fire detection algorithm for ship fire,which greatly improves the accuracy of fire smoke detection.The video-based smoke detection algorithm proposed in this paper is a cascade model.In this model,the smoke candidate area is selected first.The first step in selecting candidate areas is motion detection.Most surveillance cameras currently in use are gimbal-based rotatable cameras,and motion recognition in traditional static backgrounds is no longer applicable.This article analyzes three common methods for detecting moving objects in a static background and their advantages and disadvantages,and introduces a method for detecting moving objects in a dynamic background.In the second step,the law of smoke color detection is obtained after analyzing a large number of smoke pictures Use the above color detection model to perform color analysis on the pixels of the extracted moving foreground objects to further filter out non-smog color objects;the third step uses dark channel a priori theory to detect smoke images.All objects that meet the above three detection steps are clustered to form a smoke candidate area.In this paper,a deep learning algorithm is used for smoke final recognition,and 8-layer Convolutional Neural Networks(CNN)and DenseNet-121 models are used to train the smoke dataset.The trained smoke detection model and the candidate smoke selection are cascaded together to detect 9 collected videos,including 6 smoke videos and 3 smoke-free videos.Three indicators,namely accuracy rate,detection rate and false detection rate,are used to evaluate the detection results,and the obtained three indicators are compared with other classic video-based smoke detection algorithm evaluation results.The results show that the smoke detection algorithm proposed in this paper improves the accuracy of fire detection and reduces false alarms.It can be used in the early detection of fires to minimize the hazards of fires.