Search

In computer science, cellular neural networks (abbreviated as CNN) are known as a parallel computing model where the communication is allowed only between neighboring cells (neurons). The given research work is an extension of our publication referred as [9] with the work focused on creating a discrete-time impulsive neutral-type CNN with time-varying delays applying the method of semi-discretization. An example of such type CNN is constructed on the base of this method. Sufficient conditions for the existence of periodic solutions of the discrete-time are formulated applying the continuation theorem from the coincidence degree theory. Validation of the newly created discrete-time counterpart is done. An example of a discrete-time neutral-type CNN satisfying the sufficient conditions is given. A sigmoid function is applied for all activation functions. The weight functions are given. A periodic solution is found and the periodic table is presented.

Image segmentation in medical imaging has long been a problem in radiological image processing. Most of the image segmentation methods in traditional vision algorithms are difficult to achieve high-resolution image segmentation due to the complexity of the algorithm. This article proposes an image segmentation method based on an optimized cellular neural network. This method introduces a non-linear template and data quantization on the basis of a basic network model, which greatly reduces the computational complexity while maintaining the accuracy of image segmentation. We then applied the method to a computer-aided system to classify tumor lesions in mammograms. Finally, we propose an FPGA-based multilevel optimization architecture for energy-efficient cellular neural networks. The optimization scheme includes three levels: system level, module level, and design space. This solution improves computing performance by increasing system parallelism, using data reuse technology to fully utilize loading bandwidth, and using data quantization to reduce computational redundancy. It also introduces pipeline and dual cache structures to optimize memory access, and analyzes the limited resources through the Roofline model. System for best performance. The experimental results show that the FPGA accelerator in this article can improve unit performance by 34% compared with other existing research work. The nonlinear quantified cellular neural network proposed in this article can reduce LUT resource consumption by 74% and energy of 48.2%. Compared with the original network, in the two projection position segmentation results of the mammogram, only 1.5% and 0.6% of the accuracy loss, respectively.

This paper analyses the synchronization problem of switched cellular neural networks. The admissible edge-dependent average dwell time method is used to design switching signals. By virtue of the multiple Lyapunov function technology and Newton-Leibniz formula, a sufficient condition is established to guarantee the exponential synchronization of switched cellular neural networks. Finally, a numerical example is employed to verify the effectiveness of proposed methods.

The interval matrix method is applied to multiple delayed memristive fractional-order neural networks with fuzzy cellular concept to achieve synchronization. Along with interval matrix method, Lyapunov function method and regarding discontinuous part as uncertainties, the LMI type asymptotically and exponentially synchronization results are obtained, which are also suitable for integer-order systems. And the obtained results are showed effective by a simulation example.

The adaptive control method is applied to deal with synchronization problems for a kind of time-varying delayed memristive fractional-order neural networks with fuzzy cellular concept. Along with Lyapunov function method and fractional-order calculus, the Mittag-Leffler quasi-synchronization and exponentially quasi-synchronization results are derived under a novel adaptive controller, which are also suitable for integer-order systems. And the obtained results are showed effective by a simulation example.

The popularity of convolutional neural networks and deep learning based approaches has increased continuously in the past years. These methods has enabled the solution of various practical problems, but they still are not heavily exploited in the embedded domain, which requires low-power implementation of these architectures. Cellular neural networks can provide an analogue and power-efficient implementation of these networks which also enables the exploitation of non-Boolean, beyond CMOS elements such as memristive structures. In this paper we will demonstrate on simple datasets how cellular neural networks with different cell dynamics can be efficiently applied in the problem of semantic segmentation.

Cellular neural network is a well-known mathematical model for investigation of deterministic, stochastic and chaotic processes in nonlinear dynamic systems. The turn to a reaction-diffusion cellular neural network (RDCNN) enables to model different types of autowave processes (concentric, circular and spiral waves, Turing structures, etc.) RDCNN is a convenient architecture for the mathematical description of the systems of nonlinear differential equations with diffusion and their solution based on numerical methods. Running circular and spiral waves in RDCNN are synthesized and their properties are demonstrated (preservation of the shape and amplitude of the running waves, as well as the destruction of waves which results from their collision.) Autowaves are built in the MATLAB system on using the fourth-order Runge–Kutta numerical method.

The Convolutional Neural Networks (ConvNets) with its proper hierarchical structure are known as powerful image-recognition processing architecture. In particular, the ConvNets are well suited for several image processing tasks, such as image classification data set, computer vision and natural language processing. Nevertheless, the implementation of ConvNets requires a large amount of operations as the 2-D convolutional mappings that need a very large computational power. In this paper, the authors will investigate alternative hardware architectures, based on Cellular Neural Networks (CeNNs), in order to improve the overall performances

This work proposes a Red Green Blue (RGB) image encryption and decryption digital system which uses random number sequence generators as core and also provides its internal architectural layout. SC-CNN (State Controlled Cellular Neural Network) based chaotic system which stands optimal in generating multi-scroll chaotic generators is inculcated in this system. The hyperbolic tangent function used in state equations of the chaotic generator is implemented via CORDIC (Coordinate Rotational Digital Computer) algorithm, which is an efficient algorithm to compute various trigonometric and hyperbolic functions. The above techniques are combined together to give hardware credibility to the scheme described. The randomness and the level of encryption are analyzed and validated with the help of multiple test inputs and corresponding encrypted outputs. The complete encryption and decryption flows are simulated using Xilinx Vivado 2019.1 and realized on FPGA (Field programmable gate array), Zynq 7 board, as the chosen device.

In this work, we propose a methodology for training a Cellular Neural Network based on the Artificial Bee Colony Algorithm and the Nelder-Mead Algorithm. The results of this proposal are compared with training using only the Artificial Bee Colony algorithm and with the Nelder-Mead algorithm, also with related works. The Cellular Neural Network is directed and trained for image processing, and the selected tasks for the performance evaluation of the Cellular Neural Network are edge detection and noise remover.

The Cellular Neural Networks (CNN) was developed by Chua and Yang in 1998. The performance of the CNN depends on the parameters which are called the template. The CNN is applied to various image processing by changing the template. The output image processed by CNN is a binary image. Therefore, the unnecessary objects are removed in the process. In this research, we propose a method of switching two templates to stop the image processing in a certain state and output in the grayscale state.

Cellular neural networks were used with success in the past decades and helped laying the foundations of neural net-work applications in image processing. In the last few years convolutional networks have appeared and helped in the solution of complex practical problems. Meanwhile programming templates of cellular neural networks were designed by analytical methods, gradient based optimization is applied popularly in convolutional networks. In this paper we will demonstrate how these methods can be exploited using cellular networks and how they can be used to implement classification and feature extraction tasks, both with standard and memristive cell dynamics.

We discuss the design of an area-efficient CMOS analog core-cell implementing a PUF derived from a two-neurons Cellular Neural Network (CNN). The study is based on both theoretical modeling and numerical simulations, proposing circuit solutions in which the area consumption is strongly reduced by eliminating state capacitors and relying on distributed parasitic capacitances only.

This work presents the training and realization of a CMOS analog two-layer FSR CNN obtained through the extension of an existing one-layer version. A genetic algorithm has been developed for the learning process and has been applied to the determination of templates parameters in a gray-scale image filtering task. Networks implementing first order Butterworth spatial filters have been simulated and the results are compared to its ideal model and FFT computations, allowing to validate the proposed learning process and circuit applicability.

In this paper, a two-grid coupled Cellular Neural Networks (CNN's) is proposed, which is composed of time-delay Chua's circuit with same lossless transmission line (LTL). Under zero-flux boundary conditions, the circuit is reduced to two neutral partial differential equations with time delay and diffusion. First, the equilibrium point of the system is calculated by its initial state and original function. Secondly, select the time delay as a bifurcating parameter, the stability and Hopf bifurcation behavior of the coupled circuit system are studied near the equilibrium point. Finally, the numerical simulation shows that the length of the lossless transmission line has an effect on the dynamic behavior of the system.

Image encryption is an effective technology to protect digital image confidentiality. This paper presents an image segmentation encryption algorithm based on a hybrid chaotic system. First, a chaotic sequence is obtained by iterating a Quantum Cellular Neural Network (QCNN), and then it is scrambled by a 4-D hyperchaotic system to generate a key pool. Second, the chaotic pointers generated by 3-D chaotic systems and QCNN with different initial values are used to get the keys for image segmentation, scrambling, and diffusion from the key pool. Then, the plain-image is divided into two blocks by the chaotic segmentation method and scrambled by intra-block and inter-block pixel exchange. In addition, two blocks are statically diffused, and the cipher-image is obtained by dynamic diffusing after combining the image blocks. Especially, the key pool increases the efficiency of the proposed algorithm, and chaotic segmentation reduces the cipher-image pixel correlation. Finally, the simulation results and performance analysis indicate that the proposed algorithm has a well-security, high sensitivity, and faster speed.

In this article, we perform a uniform benchmarking for the convolutional neural network (CoNN) based on the cellular neural network (CeNN) using a variety of beyond-CMOS technologies. Representative charge-based and spintronic device technologies are implemented to enable energy-efficient CeNN related computations. To alleviate the delay and energy overheads of the fully connected layer, a hybrid spintronic CeNN-based CoNN system is proposed. It is shown that low-power FETs and spintronic devices are promising candidates to implement energy-efficient CoNNs based on CeNNs. Specifically, more than $10\times $ improvement in energy-delay product (EDP) is demonstrated for the systems using spin diffusion-based devices and tunneling FETs compared to their conventional CMOS counterparts.

Cellular neural networks have been applied in many fields and have shown good results. Therefore, this article studies the leader-following consensus problem (MSALFC) of multi-agent systems with fixed directed topology under cellular neural networks framework (CNNF). Compared with the common consensus problem, agents' communication topology is reconstructed through CNNF. Then, the breadth-first algorithm is used to cut unnecessary connections in the system. So that the undirected communication topology based on CNNF is transformed into a directed topology with only one spanning tree and communication protocol are constructed. Finally, an effective control protocol is designed to help the system achieve leader-follower consensus and Lyapunov stability theory is used to establish the sufficient condition to guarantee the realization of MSALFC. The communication protocol based on CNNF can not only effectively achieve the consistency of the system, but also reduce the communication traffic in the System. Numerical simulation is given to illustrate the theoretical results.

A cellular neural network (CNN) with a bipolar stepwise activation function is considered. The influence of the weight quantization levels number on CNN information capacity is investigated. It is shown that the capacity of CNN with quantized weights and the number of quantization levels, sufficient to approach the capacity of CNN version with continuous weights, depend upon CNN training method (Hebb method or local projection method).

In this work, we propose a Cellular Neural Network (CNN) implemented within a FPGA evaluation board for image processing tasks; efficiency and feasibility are evaluated with an architecture and an algorithm that develop global connectivity detection and as a shadow detector using $\mathbf{8}\times \mathbf{8}$ cells image array. The network has been implemented using the Time-Multiplexing algorithm, to which a modification has been proposed to perform image processing, increasing this way the algorithm efficiency while processing time and resources consumption are reduced, compare with other approaches.

This paper presents the architecture of a hardware accelerator for a cellular neural network (CeNN) with an application to real-time edge detection on visible-range and infrared video. The accelerator features fully-pipelined processing elements (PEs) that exploit the data parallelism in the algorithm to perform an iteration of the CeNN on a stream of video data with high throughput. The memory architecture exploits the locality of reference in the CeNN, so that each PE uses only 5 line buffers to store pixel, state, and output data, thus achieving low on-chip memory utilization. Implemented on a Xilinx XC7A200T FPGA running at 245MHz, the accelerator performs edge detection on 1080p video using a single CeNN iteration with a throughput of 118 frames per second (fps), a total latency of 15.7us, and 618mW of power consumption. The architecture features static reconfiguration to store built-in kernels and to add more PEs to support multiple iterations of the CeNN algorithm. More kernels can be added dynamically through a serial interface.

In this paper, sports video images are taken as the main research object, with the video processing technology, to design and perform a set of methods including video controlling, panorama synthesis, static background suppression, moving object tracking and quantification functions for the sports analysis system. Aiming at the video sequence with relatively static background, this paper also proposes a video moving object segmentation algorithm based on cellular neural network, to complete the program design and development based on MATLAB experimental software platform, and to further improve traditional algorithm by analyzing the experimental results. Finally, the experimental results show that the algorithm shows performance advantage in segmentation performance and application.

In order to improve the efficiency of the combined harvester-grain truck operation, and realize the non-stop unloading of grain, a machine vision-based method for detecting the loading status of the grain truck is introduced. We propose a edge detection model of the cellular neural network (CNN) using the ant lion optimization algorithm (ALO) to identify the edge of the grain bin, and segment the grain bin area by improved random line detection (RLD) method. The grain area is obtained by HSV color feature transformation and the grain convex hull is obtained by the convex hull algorithm. The distance from the grain hull to the edge of the grain bin is measured in real time, and a threshold is preset to determine the loading status of each part of the grain bin.

The effect of non-homogeneities in cell parameters for a Cellular Neural Network (CNN) architecture consisting of linear cells locally connected within a specified neighborhood and able to exhibit pattern formation is investigated. The results are an extension for another architecture of Turing pattern formation in nonhomogeneous two-grid CNN's and are based on a continuity theorem about the dependence of polynomial roots on their coefficients.

This article proposes the combination of two bio-inspired computational models that are sequentially implemented to eliminate impulsive noise and edge detection in grayscale images. In general, this procedure consists of: (1) implementing a cellular automaton (CA) with an adaptive behavior that expands the Moore neighborhood when he considers that the information obtained from his first level neighbors is insufficient. Based on the above, the image affected by noise is processed, in order to eliminate the corrupted pixels and perform reprocessing that will lead to the improvement of the quality of the image, (2) the resulting image is defined as an input of the cellular neural network (CNN) together with the training images, so that by defining three templates (feedback (A), cloning (B) and threshold or bias (I)), you can perform contour detection of objects present within the image thrown by the initial method. The results for the noise elimination presents a restoration of the image that oscillates between 70.63% and 99.65%, indicating that the image does not lose its quality despite being exposed to high noise levels, similarly it occurs for the edge detection, which presents an approximate efficiency of 65% with respect to the algorithms established within the framework of comparison.

As the performance of integrated circuits has improved, attention has focused on functions of the brain having the most advanced information processing capability. Therefore, attempts to realize the functions with engineering approach have been actively conducted around the world. Current mainstream of digital computers have strict, logical, serial, and concentrated information processing functions, but they do not have flexible, intuitive, parallel, and distributed functions like the human brain. Therefore, it is expected that neuromorphic computing has unprecedented innovative processing functions. The time required for information processing of brain-type computing is determined by the response time of the memory. Therefore, conventional digital computers require several tens of steps for the associated memory operations, but analog neuromorphic systems can do so in one step. In this article, we have implemented and evaluated neuromorphic hardware modeling human brain neurons using cellular neural networks and oxide semiconductor a-IGZO. A cellular neural network is used because it is suitable for large-scale integration of electronic equipment. Since oxide semiconductor elements can be used as synaptic elements by improved Hebb learning, deterioration in characteristics is thereof used as strength plasticity of synaptic connection. A letter correction system is developed by a cellular neural network and oxide semiconductor synapses, and the letter correction rate is evaluated by logic simulation.

Optimization methods took a leap forward in the past years and with the appearance of deep neural networks and gradient based methods, various complex and differentiable problems become solvable in myriads of practical tasks. Meanwhile optimization of discrete problems is still a challenging task, where efficient solvers are required. In this paper we present a novel analogue circuit in simulation, inspired by the dynamics of cellular neural networks, which can solve NP-hard optimization problems in continuous time. The dynamics were tested in MATLAB and PSPICE and it is shown via moderately complex problems that the dynamics of the circuit converges to the optimal solution in milliseconds with low power consumption.

This paper presents a comparative performance analysis of two learning algorithms developed for the use in Cellular Neural Networks (CNN): the Center of Mass Algorithm, a back-propagation like technique, and an adaptation of the Genetic Algorithm. Both methods are applied for the training of a CNN built with Full Signal Range (FSR) cells, for the implementation of several well-known bipolar functions of image processing. Performance parameters such as total execution time, number of CNN runs and success rate are assessed in order to provide guidelines for the learning method choice.

This work presents a CMOS analog multiplier architecture for application as the synapse in analog cellular neural networks. The circuit comprises two voltage-mode inputs and a current-mode output. Simulated performance features obtained from a circuit design in CMOS 130 nm technology include: +100 mV input voltage range, 23 µW static power, −32 dB maximum total harmonic distortion and −3 dB bandwidth of 51.2 kHz. The active area totalizes only 40 µm2.