Neural Network concept is based on "Learn by example". meansquareerror function is the basic performance function which affects the network directly. Reducing of such error will result in an efficient syst...
详细信息
ISBN:
(纸本)9783319029313;9783319029306
Neural Network concept is based on "Learn by example". meansquareerror function is the basic performance function which affects the network directly. Reducing of such error will result in an efficient system. The paper proposes a modified meansquared error value while training Backpropagation (BP) neural networks. The new cost function is referred as Arctan meansquareerror (AMSE). The formula computed prove that the modification of MSE is optimal in the sense of reducing the value of error for an asymptotically large number of statistically independent training data patterns. The results shows improved network with reduced error value along with increment in performance consequently.
Electrical impedance tomography (EIT) imaging method is gaining its popularity due to ease of use and also non-invasiveness. The inner distribution of resistivity, which corresponds to different resistivity properties...
详细信息
ISBN:
(纸本)9781479944125
Electrical impedance tomography (EIT) imaging method is gaining its popularity due to ease of use and also non-invasiveness. The inner distribution of resistivity, which corresponds to different resistivity properties of different tissues, is estimated from voltage potentials measured on the boundary of inspected object. The major problem of EIT is how to reconstruct the image of inner resistivity. There are many approaches to solve this issue, which require more computational demands. The use of neural network to solve this non-linear problem addresses the demand to ease the implementation and lower the computational demands. In this article we adopted the use of Radial Basis Function (RBF) neural network for image reconstruction and compared it to reconstructed images obtained using EIDORS software. RBF network was created and trained using the MatIab and neural network toolbox. As training data the simulated measurement voltages and EIDORS difference reconstruction gained values of model elements were used as input and output vectors. Then we performed testing onto 100 images and compared them with images reconstructed with EIDORS difference reconstruction. To calculate the error we used mean square error algorithm.
暂无评论