The Study On The Application Of RBF Neural Network In ECT Image Reconstruction

Two-phase flow parameters measurement is a difficulty in field of industrial process parameter measurement. Process tomography (PT) technique can provide interior behavior 2/3 dimension visualizing information in process equipment such as closed pipe, vessels and so on in real time, so it is a novel approach for two-phase flow parameters measurement. Electrical capacitance tomography (ECT) is a technique based on capacitance sensor. ECT has the advantage of being non-radiate, non-intrusive, fast in response, simple in structure and low in cost, so it is one technique which has great developing potential in industrial application.Image reconstruction algorithm is the ECT key technique. Image reconstruction that reconstruct image in the imaging area by measuring capacitance around the imaging area is a nonlinear and ill-posed inverse problem.Radial Basis Function neural network (RBFNN) is a general approaching instrument, and if only it holds enough number of nerve cell in hidden layers, can approach any multiple nonlinear continuous functions at will,and RBFNN infers from regularization theory that is good at resolve inverse problem. Hence, This dissertation discussed ECT Image Reconstruction method based RBFNN.Genetic Algorithm (GA) is a global algorithm which uses natural law "survival of the fittest, extinction of the unfitness" and natural selection and heredity mechanism to guide and determinate search direction. Comparing with other optimization methods, it needs few mathematical requirements, so it can be applied in many fields. This dissertation introduces two-layers GA as main body of study algorithm of RBFNN. In two layers GA, the first layer is to evolve the number of nerve cell in hidden layers of RBFNN, and the second layer is to evolve the nerve cell function's parameter in hidden layers. In the second layer, K-nearest neighbor algorithm is introduced to ascertain searching scope firstly, and then the nerve cell function's parameter in hidden layers begin to be evolved in this scope. The Least-Square is also introduced to calculate connection power between hidden layer and output layer. In addition, evolution's effect of two-layer GA is discussed.This dissertation conducts a series or experiments with the image reconstruction method brsed on RBFNN, compared with LBP algorithm, and the results of experiment show that the method based REFNN can reconstruct good quality image.Generalization ability of RBFNN is also analysed with experiments. In addition, an estimating parameter named area relative correct rate is presented.Lastly, this dissertation discusses detailedly the data-acquisition that is a important part of ECT system, and designs imaging software of ECT system, including real time and non-real time imaging.