Research On Theories And Key Technologies Of Power Schottky Rectifiers

Image recognition technology refers to the use of image processing, pattern recognition,intelligent optimization algorithms and computer-related technology acquisition to analysisthe collected image,and get the small amount of useful information which can be used inimage recognition.The image recognition system studied in this paper are consist of those fourmajor components,which are the choice of image preprocessing, feature extraction, featurevector value, classification and identification of four.This paper,utilized the weighted average method and the global threshold binarization toget image pre-processing, but usually dimensionality of feature vectors extracted from theimage is too high, only the value whcih can express the similar differences can be effectiveto the identification. According to this problem of the Invariant moments extraction for thewavelet feature vector dimension which is too large, in order to improve the performance ofthe classifier,this paper proposed the optimization algorithm based on improved invasiveweeds characteristic vector selection algorithm.While the algorithm is effective to solve theproblem of the choice of the eigenvectors,and effectively reduce the dimensionality of thefeature vector.Experimental simulation shows that its performance is better than featurevector selection utilizing genetic algorithms.Finally, utilized three layer BP nerve net as the ultimate classifier and also use the LMalgorithm of BP network training methods,the range of feature vectors which extracted fromwavelet invariant moments is in the range of real numbers,and is relatively large can not beused as the input of the neural network, while the initial value of weight is in the smallerrange, so this paper, a mapping function is given,which can mapped to a small area,theexperimental results show that this method reduced training frequency and training time ofthe neural network.