摘 要: 支持向量機(jī)在處理樣本類別不平衡時(shí),對(duì)樣本數(shù)量少的類別,其分類誤差大,針對(duì)這一問(wèn)題提出了一種多核心的加權(quán)向量機(jī)。將紙幣的RGB色彩空間轉(zhuǎn)換到Y(jié)IQ色彩空間,使用直方圖對(duì)樣本進(jìn)行特征的提取。在特征空間使用邊界區(qū)間處理分類問(wèn)題,加入最小化小樣本誤識(shí)率的機(jī)制,使用一種改進(jìn)的加權(quán)支持向量機(jī)來(lái)識(shí)別樣本。對(duì)樣本的不同區(qū)域使用不同的核函數(shù)。實(shí)驗(yàn)表明本算法可行、可靠。
關(guān)鍵詞: 多核學(xué)習(xí); 加權(quán)支持向量機(jī); 紙幣識(shí)別; 懲罰因子; 加權(quán)值
中圖分類號(hào):TP391 文獻(xiàn)標(biāo)志碼:A 文章編號(hào):1006-8228(2013)02-34-04
Recognition of counterfeit money based on improved weighted support vector machines
Lin Shuqing, Li Rong
(ZengCheng college of South China Normal University, Guangzhou, Guangdong 511363, China)
Abstract: To deal with the unbalanced problem in classification, a multi-kernel Support Vector Machines is presented in this paper. RGB color space of note is transformed to the YIQ color space and the characteristics are extracted by using histogram. The classification problem is dealt with by using the boundary interval in eigen-space, and an improved weighted support vector machines are added to minimize the 1 alarm rate of the small sample. The different kernel functions are used in different regions of the sample. Simulation results show that this algorithm is feasible and reliable.
Key words: multi-kernel learning; weighted support vector machines; recognition of the paper money; penalty factors; weighted value
0 引言
紙幣的防偽技術(shù)是世界各國(guó)金融系統(tǒng)所面臨的十分重要又棘手問(wèn)題。目前主要依靠全息磁性安全線磁碼、紫外光、紅外投射檢偽。但在高端產(chǎn)品中僅靠以上技術(shù)識(shí)偽是不夠的,特別是現(xiàn)在很多是使用清分機(jī)對(duì)高仿紙幣的識(shí)別。因此國(guó)外很多學(xué)者不斷地研究探索智能識(shí)別系統(tǒng),如Takeda等人利用神經(jīng)網(wǎng)絡(luò)系統(tǒng)去分析、識(shí)別紙幣的特征信號(hào)[1,2];Omatu等人利用遺傳算法分析、識(shí)別紙幣的特征信號(hào)[3];He等人將紙幣切割成m×n個(gè)等分區(qū)域,使用GA算法進(jìn)行識(shí)別[4];近年來(lái)Jae-Kang等人[5]和Ahmadi等人[6]也進(jìn)行了相應(yīng)的研究。
本文使用多核心加權(quán)向量機(jī)來(lái)識(shí)別紙幣,將紙幣圖像分割成m×n個(gè)區(qū)域,然后使用改進(jìn)的加權(quán)向量機(jī)進(jìn)行識(shí)別。對(duì)每個(gè)區(qū)域構(gòu)建多核心的核矩陣。實(shí)驗(yàn)結(jié)果證明該算法的計(jì)算精確度高,魯棒性好。
1 特征提取
假鈔與真鈔的顏色對(duì)假幣識(shí)別用處不大,針對(duì)假鈔與真鈔在色度分布上的差異,我們將RGB色彩空間轉(zhuǎn)換到Y(jié)IQ色彩空間。……