Research On Modeling Method Of Sum Product Network For Natural Scene Recognition

In the field of image scene recognition,natural scenes become a major difficulty due to their complex spatial structure and the external representation of differences between similar objects.The proposed Sum Product Network provides the researchers with an efficient probability derivation framework that can guarantee both accuracy and efficiency.The semantic region in the image is regarded as the feature subset,and the method of applying the sum product network to i:mage segmentation can be obtained.This paper first proposes a hierarchical Sum Product Network model,and designs and implements a two-layer SPN model.By modeling potential function values and categories at different scales,it is possible to adequately handle associations between different sized category regions to obtain a more accurate distribution of obj ect categories.Secondly,this paper studies the implementation details of the learning and derivation algorithms of hierarchical SPN.Experiments were performed on two natural scene datasets.The experimental results show that the hierarchical SPN can obtain more accurate scene recognition results as well as excellent robustness,and the network is more efficient than most existing models.Finally,this paper designs and implements an image acquisition system with sensor data,which can be combined with an unmanned aerial vehicle to obtain natural scene images in a real environment.At the same time,the system can obtain sensor information of the mobile device while performing image acquisition to assist image recognition.