Catálogo de publicaciones - libros

This paper proposes several improved CMOS analog circuits for neuro-fuzzy network, including Gaussian-like membership function circuit, minimization circuit, and a centroid algorithm defuzzier circuit without using division. A two-input/one-output neuro-fuzzy network composed of these circuits is implemented and testified for non-linear function approximating. HSPICE simulation results show that all the proposed circuits provide characteristics of high operation capacity, high speed, simple structures, and high precision. They are very suitable for rapid implementation of neuro-fuzzy networks.

High Quality CD, and DAT audio is very data intensive. Currently, the multi-channel technique is the preferred method of audio transmission. The MPEG(Moving Picture Experts Group) provides data compression technology for sound and image systems. The MPEG-2 AAC standard provides multi-channel 5.1 sound, using the same audio algorithm as MPEG-1, thus MPEG-2 audio both forward and backward compatible. The MDCT(Modified Discrete Cosine Transform)is a linear orthogonal lapped transform based on the concept of TDAC(Time Domain Aliasing Cancellation). In this paper, we propose an efficient algorithm for the optimization of the core in the audio part of the data transmission based on the MDCT/IMDCT(Inverse MDCT). This algorithm reduced the operating coefficient by overlapped area to bind. In the comparison of the original algorithm with the optimized algorithm that cosine coefficient reduced 0.5%, multiplies operating 0.098% and adds operating 0.58%. The proposed Algorithm was implemented using the C language then we designed hardware architecture of micro-programmed method it’s applied to optimized algorithm. This processor was designed with the VHDL language and was synthesized using the design analyzer of SYNOPSYS, with rule checking by SADAS. This processor operates at a clock frequency of 20MHz and a voltage of 5V. Thus, the designed system can be used for systems based on other FPGA and ASIC.

In this paper the neuron operation using stimulated Brillouin scattering (SBS) in optical fiber is described. The inherent optical feedback by the backscattered Stokes wave in optical fiber leads to instabilities in the form of optical chaos. At low power, the nature of the Brillouin instability can occur below threshold. At high power, the temporal evolution above SBS threshold is periodic and can become chaotic. Control of chaos induced transient instability in Brillouin-active fiber is experimentally implemented with Kerr nonlinearity having a non-instantaneous response in netowork systems. Controlling chaotic instabilities can lead to multistable periodic states; create optical logic or high level or , or low level . It can be used in neural networks. It can also lead, in principle, to large memory capacity.

Parallel genetic algorithms are usually implemented on parallel machines or distributed systems. This paper describes how fine-grained parallel genetic algorithms can be mapped to programmable graphics hardware found in commodity PC. Our approach stores chromosomes and their fitness values in texture memory on graphics card. Both fitness evaluation and genetic operations are implemented entirely with fragment programs executed on graphics processing unit in parallel. We demonstrate the effectiveness of our approach by comparing it with compatible software implementation. The presented approach allows us benefit from the advantages of parallel genetic algorithms on low-cost platform.

This study proposes the algorithm for segmentation liver and segmentation vessel inside the liver by using MDCT image. There are two main vessels in the liver. During the transplantation, it is important to decrease damage on the vessels and to raise the rate of success by providing medical doctors with the necessary incision rate of the liver and type of the vessels before operation. When transplanting, the size of donator’s liver is important for the survival of both donator and receiver. For the survival of both, the donator should leave 35% of his/her own liver, and the receiver should get more than 40% of his/her own liver. By finding out distribution of essential vessels that determine the cutting part for the transplantation and by showing artery and vein separately from the several segmentation vessel image, we can find the liver vein, which is the most important criterion during the incision, and can progress the cutting of the liver from the liver vein. It can be of help to minimize the damage on the three thick vessels and their surrounding vessels, and to cut the liver according to the volume rate of the liver. Using the features that each vessel has circle type and stick type with many angles, segmentation liver through morphological filtering and segmentation liver vessel were performed. Then, the separation of artery and vein from other combined vessels, and its reconstruction was possible, and finally the 3Dimension vessel image was produced.

In this paper, a semi-actively tuned liquid column damper (TLCD) based on fuzzy neural networks (FNN) is proposed to vibration control of irregular buildings excited by multi-dimensional earthquake ground motions. The fuzzy neural networks method takes advantage of both neural networks and fuzzy controls and has the unique combination of ability to learn via nonlinear mapping of neural nets and the capacity to integrate expert knowledge via fuzzy rules. The fuzzy neural networks based on Takagi-Sugeno model is adopted in this research to actively adjust the orifice opening-area of the TLCD. An eccentric building equipped with two TLCDs arranged in perpendicular directions is used as an object for suppressing vibrations induced by multi-dimensional earthquake ground motions. For numerical simulations, a state space representation of the building-TLCD system is derived. Numerical simulations demonstrate that TLCDs regulated by the fuzzy neural networks are effective in controlling both the translational and rotational seismic response of the eccentric building.

This paper addresses an obstacle avoidance issue for redundant nonholonomic mobile modular manipulators. On the basis of modular robot concept, an integrated dynamic modeling method is proposed, which takes both the mobile platform and the onboard modular manipulator as an integrated structure. A new obstacle avoidance algorithm is proposed which is mainly composed of two parts: a self-motion planner (SMP) and a robust adaptive neural fuzzy controller (RANFC). One important feature of this algorithm lies in that obstacles are avoided by online adjusting self-motions so that the end-effector task will not be affected unless the obstacles are just on the desired trajectory. The RANFC does not rely on exact aprior dynamic parameters and can suppress bounded external disturbance effectively. The effectiveness of the proposed algorithm is verified by simulations.

In [1], a variant version of the fuzzy cellular neural network, called FCNN, is proposed to effectively segment microscopic white blood cell images. However, when applied to the segmentation of serial CT liver images, it does not work well. In this paper, FCNN is improved to be the novel neural network —Advanced Fuzzy Cellular Neural Network AFCNN. Just like FCNN, AFCNN still keeps its convergent property and global stability. When applied to segment serial CT liver images, AFCNN has the distinctive advantage over FCNN: it can keep boundary integrity better and have better recall accuracies such that the segmented images can approximate original liver images better.

Motor unit action potentials (MUAPs) recorded during routine electromyography (EMG) examination provide important information for the assessment of neuromuscular disorders, and the neural network based MUAPs classification system has been used to enhance the diagnosis accuracy. However, the conventional neural networks methods of MUAP diagnosis are mainly based on single feature set model, and the diagnosis accuracy of which is not always satisfactory. In order to utilize multiple feature sets to improve diagnosis accuracy, a hybrid decision support system based on fusion multiple neural networks outputs is presented. Back-propagation (BP) neural network is used as single diagnosis model in every feature set, i.e. i) time domain morphological measures, ii) frequency parameters, and iii) time-frequency domain wavelet transform feature set. Then these outputs are combined by fuzzy integral. More excellent diagnosis yield indicates the potential of the proposed multiple neural networks strategies for neuromuscular disorders diagnosis.

This paper presents the implementation of adaptive fuzzy neural network controller (FNNC) for accurate speed control of a permanent magnet synchronous motor (PMSM). FNNC includes neural network controller (NC) and fuzzy logic controller (FC). It combines the capability of fuzzy reasoning in handling uncertain information and the capability of neural network in learning from processes. The initial weights and biases of the artificial neural network (ANN) are obtained by offline training method. Using the output of the fuzzy controller (FC), online training is carried out to update the weights and biases of the ANN. Several results of simulation are provided to demonstrate the effectiveness of the proposed FNNC under the occurrence of parameter variations and external disturbance.