Abnormal Detection And Visualization Analysis Based On Monitoring Data Of Earthen Ruins

Internet of Things(IOT)can monitor the status of earthen ruins in real time.Analyzing the monitoring data accurately and efficiently has important implications for the preventive protection of large earthen ruins.For earthen ruins,the abnormal situation of the monitoring data often represents the abnormal change of the conserve status,the ontological data or the diseases in the earthen ruins.Based on the abnormal detection and visual analysis of IOT monitoring data,and detailed analysis of IOT monitoring data by using machine learning methods to identify the abnormal environmental conditions of the earthen ruins.And analysis of environmental monitoring data by visualization methods provides scientific basis for preventive protection of earthen ruins.In this thesis,the specific research contents include:1.An image change detection method based on the Principal Component Analysis(PCA)and the Fuzzy C-Mean(FCM)clustering is adopted to detect the changing area of the disease in earthen ruins.The difference method and the log-likelihood ratio method are used to generate a difference image,which effectively avoids the influence of noise.Then the standard orthogonal feature vector is extracted by principal component analysis of the non-overlapping block set of the difference image,and the feature vector is mapped onto each pixel space in turn to form a feature vector space.The fuzzy C-means clustering method is used to divide the feature vector space into two clusters according to the degree of feature vector approximation.The experimental research on the image of the earthen ruins in XI'an Tang the imperial City Hangguang entrance shows that the method can efficiently analyze the changing area of the pastry disease on the wall of the earthen ruins.2.An improved nonlinear dot plots based on an undirected reassignment algorithm is proposed.The improved nonlinear dot plots can process and assign data to achieve an optimized layout which can find the suitable dot position and data distribution.The proposed algorithm is more effective in displaying dot and avoiding overlaps.Outliers dots with large differences can be shown intuitively in the nonlinear dot plots.The proposed method in this thesis can be combined with other data visualization attributions for data analysis.By drawing nonlinear dot plots with the monitoring data of the Ming Great Wall Earthen Ruins,the improved nonlinear dot plots not only allow dot size change,but also clearly display extremely data density distribution.The weight-based fuzzy twin support vector machine(FTWSVM)is used to identify the anomaly data based on earthen ruins.FTWSVM aims to generate two non-parallel planes which making each plane closer to one of these two classes and as far away from the other as possible.The different weights are reproduced on the error variables to eliminate the influence of the noisy data.The fuzzy membership degree is calculated in the feature space,and the size of the fuzzy membership degree is represented by a kernel function.This method has good performance in reducing the influence of outliers,and significantly improves the classification accuracy and generalization ability.The FTWSVM method was used for abnormal detection of the monitoring data in Hanguangmen earthen ruins diseased areas,and the results are superior to other conventional methods of comparison.