Research Based On Cellular Neural Network

With development of algorithm based on CNN and perfection of CNN chip,it has promoted a new round of upsurge for research on Neural Networks. On the other hand,due to development and perfection of CNN chip,characteristic of high-speed parallel of it makes algorithms based on CNN that the amount of data and calculation is heavy gain more real-time supports.The dissertation mainly studies on three algorithms based on r CNN. Main contents of algorithms are as follows:Firstly, algorithm of designing CNN template parameters based on adaptive genetic algorithm is presented.By improving the probability of inheritance and variation and genetic operators, it can improve the forwardness of GA;By using the qausi-excutness penalization function to compute fiteness function, it can make the algorithm rapidly converge to the global optimum and the amout of data is lesser.Secondly, algorithm of image segmentation based on acticve contours using CNN is presented. using organic combination of gradient field and GVF field as external force field for snakes. And designing the templates of EXP block and TH block.The simulation test validation for this algorithm is done and results shows that it can not only decrease high of complexity of tradition of implementation algorithms ,but also implemented by hardware parallelly.Thirdly, logic circuits are implementatd based on CNN. Status equation of a single cell acorrding to the network construntion chatacter and mathematical physics model of CNN is deduced;circuit diagrams with dissimilar inputs and parameters are drawed and then carrying on simulation using MATLAB,logic functions according to MATLAB simulation are defined;Then,carrying on simulation using EWB,simulation. results shows that circuit diagrams are right, the scopes of the input signal according to EWB simulation data are defined;Finally, designing practical circuit to debug circuit function.