Image Reconstruction Algorithm Based On SVM For Electrical Capacitance Tomography System

Electrical Capacitance Tomography (ECT) as a kind of online detection imagingtechnology develops rapidly in recent years. It has advantages of non intrusive，highresponse speed, simple structure, low cost, good security and wide application range.Due to factors such as it’s inherent structure, less independent capacitance samples,soft field, and nonlinear problem solved exist, the image reconstruction algorithm forECT system has become a key of the practical application. Support Vector Machines(SVM) as the hotspot of machine learning has better generalization ability and betterclassification accuracy. It could be a powerful means for ECT image reconstruction.The paper analyzes the structure and principles of ECT system at first, and thenanalyzes the application of SVM for ECT image reconstruction. According to SupportVector Machine has low accuracy and low training speed to deal with large scale datain ECT system, a new algorithm that combined SVM with the Rotation SymmetricPartition (RSPSVM) is presented. Twelve electrode capacitance sensor model isestablished using ANSYS, the finite element analysis software. And ECT model isestablished using the finite element analysis method. In the RSPSVM algorithm,model of ECT system is partitioned into equal area units with rotation symmetry. Thesample matrix is fully simplified. Sample matrix is segmented selectively into smallersample matrixes for training, and the smaller sample matrixes can be respectivelytrained by SVM. Then the decision function obtained can be used to classify theprediction sample. Finally, all prediction units are combined for. Experimental resultsshow that the image reconstruction using RSPSVM algorithm has higher classificationaccuracy and higher speed than using SVM alone.The paper analyzes the structure, principle and general design process of FPGA.Hardward design scheme is proposed for RSPSVM implement with FPGA, and isverified by simulation experiment. Finally, the image reconstruction result is comparedbetween the software algorithm and the hardware algorithm. Experimental results show that the hardware algorithm has hundreds times speed than the softwarealgorithm. Although the accuracy is slightly lower than the software algorithm, thehardware algorithm also meets the high precision requirement. At the same time, itcould be found that the algorithm has better classification result in single dropletflow, core flow and circulation flow patterns. A new way is provided to solve theimage reconstruction problem of ECT system.