Research On Fire Image Recognition And Parallel Monitoring Algorithm Based On Large Data

Forest is an important human resource and natural oxygen bar.It has the function of regulating climate,purifying air and keeping soil and water.It plays an important role in the balance of the ecological environment.But forest is also the most vulnerable place to fire.Once the fire happens,the loss will be irreparable.Therefore,the monitoring of forest fires has become a hot topic for scholars.With a wide range of forest monitoring and expanding the scope of monitoring the installation of cameras,resulting in massive growth monitoring data storage and processing,the pressure of the existing monitoring system is more and more difficult to deal with,to adapt to the fire image recognition algorithm for large-scale data processing is relatively small,so this paper for forest fire monitoring under the new situation of research and utilization Hadoop mass monitored data processing technology,and design suitable for parallel monitoring algorithm of fire image recognition to improve the recognition efficiency of massive fire image data and to meet a wide range of monitoring forest fire image requirements.Firstly,this paper introduces the basic theory of Hadoop platform,including the HDFS file system,MapReduce framework and operation principle,in view of the problem that the processing of large image small files on Hadoop platform will cause too many tasks to start and reduce the efficiency of the cluster,the serialization file is adopted to optimize the image data,which is verified by experiments.The optimized image data can improve the processing efficiency of Hadoop cluster significantly.Then,based on the difference between the fire flame image and the common interferer,the color feature,texture feature and Sift feature are selected as the recognition criteria,and the features are described quantitatively by descriptors such as center moment,energy and contrast.Then the fire image feature extraction under Hadoop is designed and implemented.Based on this,a parallel random weight multi feature fusion image classification algorithm is proposed,and a parallel SVM implementation scheme that is suitable for the algorithm is designed according to Hadoop principle.Through the random weight matrix,normalization and feature fusion steps to get the best SVM;Finally,the effectiveness of Hadoop platform in processing massive image data and large computation is verified through experiments.It is also proved that the proposed parallel algorithm not only takes a short time but also has a good recognition effect compared to the traditional algorithm.Forest fire monitoring based on Hadoop platform,on the one hand,solved massive storage with HDFS file system.On the other hand,parallel algorithm based on MapReduce framework also improved the recognition rate and processing efficiency of fire image,which has practical value.