本文提出了一種改進(jìn)的神經(jīng)網(wǎng)絡(luò)板形模式識(shí)別方法，該方法基于支持向量機(jī)（SVM）
與徑向基（RBF）網(wǎng)絡(luò)的結(jié)構(gòu)等價(jià)性，利用SVM的回歸確定RBF網(wǎng)絡(luò)較優(yōu)的初始參數(shù)，解決了傳統(tǒng)神經(jīng)網(wǎng)絡(luò)模式識(shí)別方法存在的網(wǎng)絡(luò)學(xué)習(xí)時(shí)間長(zhǎng)，易陷入局部極小值等問題。此外，由于板形標(biāo)準(zhǔn)模式具有兩兩互反性，將輸入樣本與基本模式的模糊距離差作為網(wǎng)絡(luò)輸入，使輸入節(jié)點(diǎn)減少一半，近一步實(shí)現(xiàn)了網(wǎng)絡(luò)結(jié)構(gòu)的固定化和簡(jiǎn)單化。實(shí)驗(yàn)表明，它提高了板形識(shí)別精度和速度，可推廣到其他標(biāo)準(zhǔn)模式具有兩兩互反性的模式識(shí)別中。
關(guān)鍵詞: 板形模式；識(shí)別方法；向量機(jī)；徑向基
Abstract: The Improved approach has been proposed based on the structural equivalence of
radial basis function (RBF) network and Support Vector Machines (SVM). The optimal initial
parameters of RBF network were gained through SVM regression, which has solved problems of the traditional method known as neural network with slow convergence and local minimum etc. Moreover, according to the reciprocal characteristic of every two typical patterns, the deduction of fuzzy distance measure was applied, which has got the numbers of the inputs declined by a half and developed the realization of the changeless and simple structure of the neural network. The improved RBF network approach to flatness pattern recognition based on SVM learning has been proved with high precision and speed. It could also be put into other fields in which Reciprocal
polynomials for every two typical patterns are existed.
Key words: Flatness Pattern; Recognition Approach; Vector Machines; Radial Basis Function