Research On Image Reconstruction Algorithm For ECT System

The electrical capacitance tomography technique can be regarded as a kind ofprocess tomography technique which was most widely studied at present. It hasmany distinct advantages such as no invasion, no touch and low cost. We can use it ifeach phase had different dielectric constant. As one core techology of ECT system,the image reconstruction algorithm is an important factor to improve the quality ofreconstructed image. Neural network approach is widely used in ECT imagereconstruction because of its advantages, such as parallel processing, distributedstorage, self-learning adaptive and highly fault-tolerant ability. The research ofreconstruction algorithm in this paper is based on the optimized transducers’structure parameter. The study we have completed is as follows.First of all, three major components of the ECT system are introduced:capacitive transducer, capacitance data acquisition system and image reconstructioncomputer. The operation principle of ECT system is detailed analysed in theory. Thecharacteristics of ECT system is obtained by establishing mathematical models, thenthe superiority of applying neural network to ECT image reconstruction is lead out.Secondly, more profound study is focused on the several kinds of typicalreconstruction algorithms. We know the scope of application and weakness of allkinds of reconstruction algorithms by analysing back projection. The advantages ofapplying Chebyshev neural network to ECT image reconstruction are obtained bycomparison.Once again, according to the problem of soft field characteristic and ill-posedcharacteristic of ECT system, a new image reconstruction method which usedChebyshev polynomial as the activation function is proposed in this paper. In orderto reduce the scale of network, we divide the whole network into six sub-systemsaccording to the features of sensitivity distribution. The capacitance is caculated byusing ANSYS sofware, then we adjust sample input interval by unipolar S function. The improved k-means algorithm is adopted on clustering capacitance afteradjustment, characteristic samples are selected which greatly reduced the scales ofthe training sample. We traine the data by using Chebyshev neural network and thenDelta rule is used to adjust weight.Finally, according to the selected capacitance sensor structure parameters, weset up different dielectric constant and different simulation flow. Then the imagereconstruction is compared by using different algorithms. The result show that thealgorithm not only improve the imaging stability and speed, but also greatly improvethe quality of imaging.