14945

13451

كارشناسي ارشد

هوش مصنوعي و رباتيكز

اصفهان : دانشگاه صنعتي اصفهان

1398

سيزده، 73ص. :مصور، جدول، نمودار

عبدالرضا مبرزايي

مهران صفاياني

بهزاد نظري، فرزانه شايق

1398/05/28

كتابنامه

مهندسي كامپيوتر

مهندسي برق و كامپيوتر

1398/05/28

2552235

Abstract Today a large amount of data with multiple feature is produced by various sources Analyzing and extracting information from this volume of data requires automatedprocessing methods that can be used to make the decision more accurate One of theapplications of data analysis is the detection of unusual and abnormal events Anomalydetection in image is one of the most important task of computer vision using a variety oftypes such as public security health monitoring and intrusion detection Analysis anddetection of anomalies is important because they provide useful information about theattributes and specifications of the data production process Despite the differentapplications anomaly detection is not a well defined problem The usual definition of anabnormality is a low probability event and is significantly different from other data Thegoal of this problem is estimate distribution of input dataset and in the future can find datathat has not been generated from this distribution Therefore usually an unsupervisedlearning problem is considered The thesis focuses on unsupervised learning to detectabnormalities The proposed approach is based on the auto encoders in conjunction withGenerative Adversarial Networks This method is called Adversarial Auto encoders and isa probability auto encoder that attempts to match the aggregated posterior distribution ofan auto encoder s hidden vector with an arbitrary prior distribution The proposed methodencoder is a capsule network A capsule is similar to a neuron except that the input andoutput of a capsule is a vector instead of a scalar This design enables the capsule not onlylearn a specific feature in the input image but also recognizes the angle and view point anddeformation of it The method of training capsules is a new approach called routing byagreement Routing by agreement finds the exit path from the low level capsule s to thecorresponding capsule s at a higher level With this approach there is a similarity betweenthe input and output of a capsule While in previous encoder structures have been usedconvolutional networks instead of capsule networks Applying capsule encoder instead ofconvolutional does not have problems such as the angle and view point images especiallyfor medical images The reconstruction error of learned auto encoder for normal events islow and for abnormal events is high This error in the auto encoder with a capsule encoderhad a better performance than an auto encoder with a convolutional encoder can be used toimage anomaly detection Also the adversarial error of the proposed method compared tothe adversarial error of previous methods in anomaly detection is better The resultsobtained from the implementation of the proposed structure show that the finalperformance can be increased up to 10 which indicates the effect of designing theproposed model Keywords anomaly detection adversarial auto encoder networks capsule nets unsupervised learning

عبدالرضا مبرزايي

مهران صفاياني

بهزاد نظري، فرزانه شايق