عنوان البحث(Papers / Research Title)
Hybriding Fuzzy Logic with wavelets
الناشر \ المحرر \ الكاتب (Author / Editor / Publisher)
بشرى حسين عليوي الواوي
Citation Information
بشرى,حسين,عليوي,الواوي ,Hybriding Fuzzy Logic with wavelets , Time 5/16/2011 8:58:49 AM : كلية التربية للعلوم الصرفة
وصف الابستركت (Abstract)
Most works are on wavenets (wavelets with Neural Networks NNs.
الوصف الكامل (Full Abstract)
Hybriding Fuzzy Logic with Wavenets
Abstract
These days the technologies are going toward Hybrid systems between techniques, that to compensate the imperfects and problems that may be occurred when using just one technique .A key word of this research is to use a Hybrid method that combine a wavelet theory with neural networks and fuzzy logic, through suggest an activation function as a summation of wavelet function and membership function, these functions are continuous. The chosen wavelet function is a Mexican hat wavelet ,while the membership function is a Gaussian membership function .
This method suggests new way to take decision through compute the output at closed interval of actual output values for inputs values to a network which are a files of features of images used as training set ,which is provide a wide support to work on , and then to accomplish more convergence pattern(s) ,since the suggested method performed on a pattern recognition problem. The suggested method performed on a set of images for a geometric curves Hand drawing for eights categories that gradual in complexity ,that to proving the network ability .Each category has six (in some experiments used seven) figures ,one a standard figure that characterize the category and five(six) from its fuzzy (vague) versions .
The idea of this suggested method is novel since; that it depends on recognize a pattern(s) through the closed interval for actual output of a pattern(s) that wanted to be recognized must at least be with in this interval outputs values for patterns in that category .Each category has a closed interval for its patterns outputs values different on all other categories intervals,and these intervals must not intersect with other at all, also these closed intervals can be regard as decision regions for convergence. While the regions that be between these intervals or out of the space of intervals are regarded as error regions, the proposed network try to as could as to minimize regions between these closed intervals of that categories, and then minimize errors probabilities though training .
تحميل الملف المرفق Download Attached File
|
|