Study On Target Recognition Based On Cellular Neural Networks

At present, the application of CNN depends largely on its dynamic behavior, suchas image processing, pattern recognition and applications in control field often requirthe network converges to a stable equilibrium point. Eight templates in the algorithmare all iterated to a stable equilibrium point before output the images. Videosegmentation need to compute large amount of data,so it is difficult to achievereal-time requirements.However, real-time signal processing capability and localconnection features of CNN can solve this problem effectively. CNN’s algorithm is aneffective method to solve the real-time of video data processing. Three aspectsinnovation in the paper are as follows:1. Based on Mengshukai’s CNN difference image merge video segmentationalgorithm, through the addition of noise removal and the replacement of edgedetection templates,we get an improved algorithm. This algorithm is aimed atrelatively static for background video sequences. Compared with the originalalgorithm, the algorithm takes noise interference in the image processing into accout,and the noise removal template reduced the effect of noise, also improves the peaksignal to noise ratio(PSNR) of algorithm. Improved edge detection template not onlyget a more accurate and detailed edge image，but also get more motion information, sothat the moving object segmentation has better accuracy and semantic.2. Noise is a common problem in acquisition,transmission and processing ofimage,which will decrease image quality seriously.So in image processing,if we donot reduce influence on images beforehand,the obtained results will be far from ourexpectation.According to the uniqueness and global asymptotical stabile equilibriumpoint of CNN,combining the property of saturation nonlinearity,the noise reductiontemplate is trained by giving a certain pair of input and output,finally the design canbe characterized in terms of LMIs. LMI problems are solved by MATLAB, and weget the noise removal template.Next it is proved by experiments that the de-noisingeffect is better.Compared with classical filtering operators, the template in the removalof low noise has the certain superiority.3. Edge in an image is the connected boundary between two different regions.Edgedetection is one of the most important and difficult steps in image processing andpattern recognition systems.The performance of the tasks after the edge detection,suchas image segmentation,object recognition and classification,and image registration are dependent on the information on the edge.We design the edge detection of noisyimages by employing CNNs and LMI,by using a simulation image and acorresponding ideal edge image as the input and output images,finally we calculateLMIs by Matlab to get template parameters.The experiment results showthat,compared with classical edge detection operator,the design of CNN edgedetection template could extract edge information detailed and comprehensive in thispaper.In this paper, we improve an algorithm aiming at relatively static for backgroundvideo sequences,and the algorithm is based on video moving object segmentation.Through designing programs based on Matlab7.0,therefore,experimental results showthat,the PSNR of extracted moving objects are improved.