Situation Assessment Model for UAV Disaster Relief in the City

Abstract

A situation assessment model based on structure- variable discrete dynamic Bayesian network (SVDDBN) of sorting information is proposed for Unmanned Aerial Vehicle (UAV), which can be applied in the condition of disaster relief in the city when pop-up threats appear. The model is built on the basis of the SVDDBN, makes an uncertain classification on pop- up threats observing information with the help of the posterior probability support vector machine (PPSVM), and finally inputs the classification results into the assessment model as the evidence. For the features of the multi-hidden nodes of the assessment model, the forward algorithm is introduced into the probability inference of the network model, and the inference algorithm of the SVDDBN under the multi-nodes is worked out. The situation that the UAV detects the pop-up threats in the air while conducting the disaster relief in the city is set as the background to verify the correctness of the establishment of the model and the related algorithm. formed based on this method is unchangeable, and it can not conduct the situation assessment under the pop-up threats. Focusing on such situation, in this paper, we propose to use the structure-variable discrete dynamic Bayesian network (SVDDBN) that is based on the information classification, to settle the problem of the situation assessing when the UAV suffers from the pop-up threats. The method is mainly consisted of three parts: the situation assessment model that based on SVDDBN, the uncertain classification of the observing information, and network probability inference. The SVDDBN based situation assessment model hierarchically incorporates the factors of the threat level assessment, the target value evaluation, etc. on the basis of the target recognition. According to the environment situation, it can change the partial structure of the model and the parameter; uncertain classification of the observing information chiefly adopt the Posterior Probability Support Vector Machine algorithm to deal with the problem of network evidence inputting, especially when the pop-up threat arises, the PPSVM can complete uncertain classification of the observing information by practicing a bit of observation(7), which avoid building the membership function for the different sorts of data; the network probability inference is based on the direct inference of Dynamic Bayesian Network (DBN)(8), then introduces the forward algorithm to solve the SVDDBN probability inference problem under the multi-hidden nodes, and work out the variation law of the rational formula by simplifying method to establish SVDDBN probability inference model.