倪世明 白云龍 蔣益群
摘要: 青年女性臂部體型包含了大量的非線性特征,單一的BP神經(jīng)網(wǎng)絡(luò)很難達(dá)到理想預(yù)測(cè)精度,為快速準(zhǔn)確地識(shí)別青年女性臂部體型,提高預(yù)測(cè)精度,本文構(gòu)建了一種基于思維進(jìn)化算法(MEA)優(yōu)化BP神經(jīng)網(wǎng)絡(luò)的青年女性臂部體型識(shí)別模型。首先,通過(guò)[TC]2三維人體測(cè)量獲取611名青年女性的臂部數(shù)據(jù);其次,通過(guò)主成分因子分析得到影響青年女性臂部體型特征的5大形態(tài)因子,選取5個(gè)特征指標(biāo)采用兩步聚類法將臂部體型分為4類;最后使用Matlab軟件構(gòu)建基于MEA-BP神經(jīng)網(wǎng)絡(luò)的青年女性臂部體型識(shí)別模型。實(shí)驗(yàn)結(jié)果顯示:該模型能有效識(shí)別臂部體型,準(zhǔn)確率為95.45%,與單一BP神經(jīng)網(wǎng)絡(luò)和GA-BP神經(jīng)網(wǎng)絡(luò)對(duì)比,該模型具有更高的預(yù)測(cè)精度和更優(yōu)的非線性映射能力。
關(guān)鍵詞: 青年女性;臂部體型;體型分類;MEA-BP神經(jīng)網(wǎng)絡(luò);識(shí)別模型
中圖分類號(hào): TS941.17文獻(xiàn)標(biāo)志碼: A文章編號(hào): 10017003(2022)05004208
引用頁(yè)碼: 051107DOI: 10.3969/j.issn.1001-7003.2022.05.007
在服裝結(jié)構(gòu)設(shè)計(jì)中,衣袖是重要的服裝部件,其形態(tài)和尺寸來(lái)源于人體臂部。近年來(lái),為了提高衣袖的合體性和舒適性,使衣袖結(jié)構(gòu)更符合人體臂部形態(tài)特征,許多學(xué)者對(duì)臂部體型做了相關(guān)研究。賀新等[1]對(duì)青年女性上肢形態(tài)特征進(jìn)行主成分分析,提取特征因子,將4個(gè)典型指標(biāo)分別細(xì)分為3類,定量描述了34類上肢形態(tài)差異;張雪云[2]對(duì)青年女性肩臂部特征參數(shù)進(jìn)行逐步聚類,最終得到7類肩臂部特征形態(tài);舒?zhèn)3]將老年女性的6類全臂長(zhǎng)、5類上臂圍進(jìn)行交叉組合,最終得到6類臂部形態(tài),并確定老年女性臂部號(hào)型規(guī)格;劉國(guó)聯(lián)等[4]通過(guò)拍攝青年女性的臂部照片并提取相關(guān)尺寸,建立了以臂寬為自變量的衣袖設(shè)計(jì)回歸方程。學(xué)者們通過(guò)提取人體臂部典型指標(biāo),并基于典型指標(biāo)分別細(xì)分及交叉組合,獲得臂部復(fù)合體型,體型細(xì)分?jǐn)?shù)量較多,雖然對(duì)服裝結(jié)構(gòu)設(shè)計(jì)和號(hào)型歸檔有一定的理論依據(jù),但缺乏實(shí)際生產(chǎn)中消費(fèi)者衣袖穿著舒適性需求和服裝工業(yè)批量化生產(chǎn)需求的平衡,缺少對(duì)臂部體型整體分類和識(shí)別相關(guān)的研究。
人體體型數(shù)據(jù)存在復(fù)雜的非線性關(guān)系,在體型分類識(shí)別的方法上,學(xué)者們應(yīng)用較多的有傳統(tǒng)的數(shù)理統(tǒng)計(jì)法[5-6],也有近幾年機(jī)器學(xué)習(xí)界比較熱門的神經(jīng)網(wǎng)絡(luò),尤其是BP神經(jīng)網(wǎng)絡(luò),在人體工學(xué)領(lǐng)域應(yīng)用較多[7]。BP神經(jīng)網(wǎng)絡(luò)是一種多層前饋神經(jīng)網(wǎng)絡(luò),具有原理簡(jiǎn)單、模型易搭建、能夠逼近任意非線性曲線的優(yōu)點(diǎn),已用于人體頭肩部體型識(shí)別[8]、人臉外觀預(yù)測(cè)[9]、人臉識(shí)別[10]等。但由于其初始閾值和權(quán)值是隨機(jī)選擇的,導(dǎo)致在處理人體體型復(fù)雜的非線性關(guān)系時(shí),出現(xiàn)收斂速度慢、模型易發(fā)生震蕩、陷入局部最小等問(wèn)題[10-11]。因此,需要尋求一種有效算法來(lái)優(yōu)化BP神經(jīng)網(wǎng)絡(luò)的權(quán)值和閾值,避免單一BP神經(jīng)網(wǎng)絡(luò)模型不穩(wěn)定、精度低等問(wèn)題,從而提高網(wǎng)絡(luò)識(shí)別精度。思維進(jìn)化算法(Mind Evolutionary Algorithm,MEA)是一種全局優(yōu)化算法,通過(guò)多次“趨同”與“異化”操作,可以優(yōu)化BP神經(jīng)網(wǎng)絡(luò)的初始權(quán)值和閾值,從而提高模型的穩(wěn)定性和預(yù)測(cè)精度。MEA-BP神經(jīng)網(wǎng)絡(luò)在衛(wèi)星鐘差預(yù)報(bào)[12]、竊電識(shí)別[13]、遙感影像分類[14]、土壤養(yǎng)分分級(jí)評(píng)價(jià)[15]等領(lǐng)域都有較好的應(yīng)用。因此,本文嘗試將思維進(jìn)化算法(MEA)優(yōu)化BP神經(jīng)網(wǎng)絡(luò)引入人體工學(xué)領(lǐng)域,用于青年女性臂部體型的識(shí)別。
本文以18~25歲的青年女性為研究對(duì)象,通過(guò)[TC]2三維人體測(cè)量獲取青年女性臂部的尺寸數(shù)據(jù),應(yīng)用主成分因子法分析得到臂部體型形態(tài)因子,并選取特征指標(biāo),隨后采用兩步聚類法進(jìn)行臂部體型分類。最后,在此基礎(chǔ)上構(gòu)建基于MEA-BP神經(jīng)網(wǎng)絡(luò)的青年女性臂部體型識(shí)別模型,一方面為人體體型識(shí)別領(lǐng)域提供一種新思路,另一方面也為衣袖部位結(jié)構(gòu)設(shè)計(jì)提供參考依據(jù)。
1數(shù)據(jù)收集及預(yù)處理
1.1實(shí)驗(yàn)對(duì)象與儀器
為了使臂部體型分類和識(shí)別更具針對(duì)性,本文選取了18~25歲的611名青年女性作為實(shí)驗(yàn)對(duì)象;身高150~180 cm,平均身高160.7 cm;體重40~69 kg,平均體重50.4 kg。
本文采用[TC]2三維人體測(cè)量?jī)x(美國(guó)[TC]2公司)、人體測(cè)高儀(日本Martin)及體重計(jì)(江蘇蘇宏醫(yī)療器械有限公司)等作為實(shí)驗(yàn)儀器。將測(cè)量環(huán)境溫度設(shè)置為(27±3) ℃,相對(duì)濕度設(shè)置為(60±10)%,測(cè)量環(huán)境符合裸體測(cè)量標(biāo)準(zhǔn)。測(cè)量方法參照GB/T 23698—2009《三維掃描人體測(cè)量方法的一般要求》,每一個(gè)實(shí)驗(yàn)對(duì)象穿著儀器規(guī)定服裝,重復(fù)測(cè)量3次,取平均值,以減小儀器測(cè)量誤差。
1.2測(cè)量項(xiàng)目
本文將青年女性臂部作為體型研究部位,根據(jù)服裝結(jié)構(gòu)設(shè)計(jì)的需要,參照GB/T 5703—2010《用于技術(shù)設(shè)計(jì)的人體測(cè)量基礎(chǔ)變量》,確定能夠反映臂部體型的12項(xiàng)測(cè)量項(xiàng)目和計(jì)算得到的2個(gè)派生變量,合計(jì)14個(gè)指標(biāo)。考慮到人體的對(duì)稱性,同時(shí)參照GB/T 16160—2017《服裝用人體測(cè)量的尺寸定義與方法》主要測(cè)量人體右臂,因此本文只分析人體右臂體型,測(cè)量項(xiàng)目和測(cè)量方式如表1所示。
1.3數(shù)據(jù)預(yù)處理
使用SPSS軟件進(jìn)行數(shù)據(jù)預(yù)處理,用Q-Q概率圖對(duì)臂部數(shù)據(jù)進(jìn)行正態(tài)檢驗(yàn),以全臂長(zhǎng)為例,檢驗(yàn)結(jié)果如圖1所示。數(shù)據(jù)大致落在一條斜線上,表明基本服從正態(tài)分布。經(jīng)檢驗(yàn),14個(gè)指標(biāo)均服從或近似服從正態(tài)分布。然后使用莖葉圖和箱形圖尋找異常值,全臂長(zhǎng)的箱型圖如圖2所示,SPSS軟件自動(dòng)標(biāo)記了1.5倍至3倍四分位距(Inter Quartile Range,IQR)的溫和異常值和超過(guò)3倍IQR的極端異常值[16]。通過(guò)查看并分析異常值的原始三維掃描數(shù)據(jù),綜合判斷數(shù)據(jù)的有效性,剔除極端異常樣本,最終確定有效樣本586個(gè),有效率95.91%。
2臂部體型分類
2.1描述性統(tǒng)計(jì)分析
使用SPSS軟件對(duì)青年女性臂部14項(xiàng)指標(biāo)進(jìn)行描述性統(tǒng)計(jì)分析,得到各指標(biāo)的極小值、極大值、均值、標(biāo)準(zhǔn)差、偏度、峰度等基本統(tǒng)計(jì)量,結(jié)果如表2所示,反映了臂部數(shù)據(jù)的集中趨勢(shì)、離散程度和總體分布等[17]。其中,極小值、極大值反映臂部數(shù)據(jù)的變異范圍;均值反映臂部數(shù)據(jù)的中心位置,描述集中趨勢(shì);標(biāo)準(zhǔn)差描述臂部數(shù)據(jù)的離散程度,標(biāo)準(zhǔn)差越大,數(shù)據(jù)離散程度越大;偏度和峰度描述臂部數(shù)據(jù)的總體分布情況,偏度反映數(shù)據(jù)分布的對(duì)稱程度,等于0表示對(duì)稱分布,大于0表示右偏,小于0表示左偏;峰度反映數(shù)據(jù)分布的集中趨勢(shì)高低特征,大于0表示分布較陡,小于0表示分布平緩。
由表2可知,全臂長(zhǎng)變化范圍是[39.89,70.87],是臂部指標(biāo)中變異程度最大的;其標(biāo)準(zhǔn)差為4.63,表明全臂長(zhǎng)變量與均值之間差異較大,波動(dòng)最大;其偏度為0.23,峰度為-0.13,表明全臂長(zhǎng)為平緩的右偏態(tài)分布。臂根圍標(biāo)準(zhǔn)差為3.11,說(shuō)明樣本之間存在較大的差異性;其偏度為0.33,峰度為0.96,表明臂根圍為陡峭的右偏態(tài)分布。由此可見,臂部體型的差異主要體現(xiàn)在長(zhǎng)度和圍度方面。
2.2主成分因子分析
利用SPSS軟件對(duì)14項(xiàng)指標(biāo)進(jìn)行主成分因子分析,通過(guò)分析各主成分方差貢獻(xiàn)率、累計(jì)貢獻(xiàn)率,對(duì)特征根大于1的前5個(gè)主成分進(jìn)行提取,累計(jì)方差貢獻(xiàn)率為88.94%,如表3所示,可以用于描述青年女性臂部體型特征。
根據(jù)最大方差法對(duì)因子載荷矩陣進(jìn)行選擇,旋轉(zhuǎn)后的成分矩陣如表4所示。比較旋轉(zhuǎn)后的載荷數(shù)絕對(duì)值,分析各成分包含的變量因子和共性,對(duì)主成分因子進(jìn)行定義。由表4可知,主成分因子1在上臂圍、肘圍、上臂最大圍、前臂圍、腕圍等圍度相關(guān)變量上載荷較大,定義為臂部圍度因子;主成分因子2在臂根圍、臂根高、袖山高等臂根部變量上載荷較大,定義為臂根因子;主成分因子3在臂根扁平率、臂根厚等臂根形態(tài)變量上載荷較大,定義為臂根形態(tài)因子;主成分因子4在全臂長(zhǎng)、上臂長(zhǎng)、前臂長(zhǎng)等長(zhǎng)度相關(guān)變量上載荷較大,定義為臂部長(zhǎng)度因子;主成分因子5在上臂長(zhǎng)比前臂長(zhǎng)上有較大載荷,定義為臂部比例因子。綜上,通過(guò)主成分因子分析得到影響人體臂部體型特征的5大因子:臂部圍度因子、臂根因子、臂根形態(tài)因子、臂部長(zhǎng)度因子、臂部比例因子。
2.3臂部體型分類
青年女性臂部數(shù)據(jù)具有非線性、數(shù)據(jù)復(fù)雜等特點(diǎn),參照王軍等[18]分類青年女性腰臀部體型所應(yīng)用的兩步聚類法,進(jìn)行臂部體型的探索性分類。兩步聚類法是一個(gè)探索性分析工具,可以揭示臂部體型數(shù)據(jù)的分類,系統(tǒng)自動(dòng)選擇最佳聚類數(shù)[17]。從表4可以看出,5大因子載荷最大值所對(duì)應(yīng)的特征因子分別是:上臂圍、臂根圍、臂根扁平率、全臂長(zhǎng)、上臂長(zhǎng)/前臂長(zhǎng),因此選取這5個(gè)變量作為臂部體型分類的特征指標(biāo)。在SPSS軟件中應(yīng)用兩步聚類法,對(duì)實(shí)驗(yàn)獲取的586個(gè)樣本進(jìn)行探索性聚類分析,發(fā)現(xiàn)青年女性臂部體型分為4類時(shí),聚類效果最佳。聚類分布情況如表5所示,特征變量平均值如表6所示。
從表6可以看出,4類青年女性臂部體型存在明顯區(qū)別。第1類臂部體型:長(zhǎng)胖臂,手臂粗壯,臂根厚實(shí),長(zhǎng)臂,上臂長(zhǎng)于前臂;第2類臂部體型:中間臂,手臂中等粗細(xì),臂根圓潤(rùn),中長(zhǎng)臂,上臂明顯長(zhǎng)于前臂;第3類臂部體型:長(zhǎng)瘦臂,手臂較細(xì),臂根勻稱,中長(zhǎng)臂,上臂長(zhǎng)于前臂;第4類臂部體型:短扁臂,手臂纖細(xì),臂根扁平,短臂,上臂明顯長(zhǎng)于前臂。
3基于MEA-BP的臂部體型識(shí)別模型構(gòu)建
3.1算法原理
3.1.1BP神經(jīng)網(wǎng)絡(luò)
BP神經(jīng)網(wǎng)絡(luò)是一種多層前饋神經(jīng)網(wǎng)絡(luò),由輸入層、隱含層、輸出層構(gòu)成[19],具有信號(hào)前向傳遞、誤差反向傳播的特點(diǎn),拓?fù)浣Y(jié)構(gòu)如圖3所示。wij、wjk為BP神經(jīng)網(wǎng)絡(luò)的權(quán)值,根據(jù)預(yù)測(cè)誤差反向調(diào)整權(quán)值和閾值,直至逼近期望輸出,在處理非線性關(guān)系上應(yīng)用廣泛。
3.1.2MEA原理
思維進(jìn)化算法是由孫承意等[20]于1998年針對(duì)進(jìn)化算法(Evolutionary Computation,EC)存在的早熟、收斂速度慢等問(wèn)題提出的一種優(yōu)化算法。MEA沿襲了遺傳算法(Genetic Algorithms,GA)關(guān)于“群體”“個(gè)體”“環(huán)境”和“進(jìn)化”等概念,并提出“群體和自群體”“公告板”“趨同”和“異化”等新概念[12]。MEA算法的趨同和異化操作,避免了GA算法交叉與變異的雙重性問(wèn)題,朝著有利方向進(jìn)化,跟遺傳算法優(yōu)化速度相比,MEA算法優(yōu)化效率更高,實(shí)用性更廣[21]。
3.1.3MEA優(yōu)化BP神經(jīng)網(wǎng)絡(luò)
在BP神經(jīng)網(wǎng)絡(luò)訓(xùn)練的過(guò)程中,由于網(wǎng)絡(luò)初始化隨機(jī)選擇權(quán)值和閾值,容易導(dǎo)致模型結(jié)果異常[11]。本文使用MEA算法優(yōu)化BP神經(jīng)網(wǎng)絡(luò),經(jīng)過(guò)趨同、異化操作,算法不斷迭代,產(chǎn)生全局最優(yōu)個(gè)體,并通過(guò)編碼規(guī)則解析最優(yōu)個(gè)體,將其作為BP神經(jīng)網(wǎng)絡(luò)初始化訓(xùn)練的權(quán)值和閾值,進(jìn)而進(jìn)行仿真預(yù)測(cè)。MEA優(yōu)化BP神經(jīng)網(wǎng)絡(luò)流程如圖4所示。
3.2MEA-BP識(shí)別模型構(gòu)建
本文根據(jù)MEA優(yōu)化BP神經(jīng)網(wǎng)絡(luò)流程,使用Matlab軟件構(gòu)建MEA-BP模型,用來(lái)識(shí)別青年女性臂部體型。
3.2.1樣本與變量選擇
實(shí)驗(yàn)中共有586個(gè)有效樣本,按照7︰3的比例分配訓(xùn)練集和測(cè)試集,其中訓(xùn)練樣本410個(gè),測(cè)試樣本176個(gè)。將主成分因子分析中提取的5大因子對(duì)應(yīng)載荷最大的5個(gè)變量(上臂圍、臂根圍、臂根扁平率、全臂長(zhǎng)、上臂長(zhǎng)/前臂長(zhǎng))作為MEA-BP臂部體型識(shí)別模型的輸入變量,把4種臂部類型(1長(zhǎng)胖臂、2中間臂、3長(zhǎng)瘦臂、4短扁臂)作為輸出變量。
3.2.2模型參數(shù)設(shè)置
構(gòu)建3層結(jié)構(gòu)的BP神經(jīng)網(wǎng)絡(luò),隱含層神經(jīng)元數(shù)量按照Hecht-Nielsen法[22]確定n=2N+1(n為隱含層節(jié)點(diǎn)數(shù),N為輸入節(jié)點(diǎn)數(shù)),計(jì)算得出隱含層節(jié)點(diǎn)為11。最終模型為MEA-BP(5-11-1)。BP神經(jīng)網(wǎng)絡(luò)模型參數(shù)設(shè)置如表7所示,MEA算法參數(shù)設(shè)置如表8所示。
經(jīng)過(guò)若干次趨同操作,MEA-BP模型達(dá)到最優(yōu)。最終優(yōu)勝子種群和臨時(shí)子種群的趨同過(guò)程如圖5—圖6所示,其中得分為模型訓(xùn)練集均方根誤差的倒數(shù)[23],分析得到:1) 各子種群得分不再增加,代表子種群均已成熟;2) 優(yōu)勝子種群1、2、3、4、5和臨時(shí)子種群1、2、3、5都沒(méi)有執(zhí)行趨同操作,該現(xiàn)象表示在這些子種群周圍沒(méi)有發(fā)現(xiàn)更好的個(gè)體;3) 臨時(shí)子種群的最高得分是子種群1,低于優(yōu)勝子種群的最低得分子種群1,此時(shí)模型達(dá)到全局最優(yōu)解,不需要再執(zhí)行異化操作,繼而解碼[23]得到優(yōu)化后的BP神經(jīng)網(wǎng)絡(luò)的初始權(quán)值和閾值。
3.3結(jié)果與分析
首先建立BP神經(jīng)網(wǎng)絡(luò)預(yù)測(cè)模型,然后引入MEA優(yōu)化BP神經(jīng)網(wǎng)絡(luò)的權(quán)值和閾值,建立MEA-BP神經(jīng)網(wǎng)絡(luò)預(yù)測(cè)模型。為了驗(yàn)證MEA-BP識(shí)別模型在人體體型識(shí)別上更具優(yōu)勢(shì),將單一BP神經(jīng)網(wǎng)絡(luò)、GA-BP神經(jīng)網(wǎng)絡(luò)、MEA-BP神經(jīng)網(wǎng)絡(luò)分別構(gòu)建青年女性臂部體型識(shí)別模型,并對(duì)比實(shí)驗(yàn)結(jié)果。三種模型預(yù)測(cè)結(jié)果、預(yù)測(cè)誤差對(duì)比如圖7—圖8所示,臂部體型識(shí)別準(zhǔn)確率如表9所示。
綜合分析可知,基于單一BP神經(jīng)網(wǎng)絡(luò)、GA-BP神經(jīng)網(wǎng)絡(luò)、MEA-BP神經(jīng)網(wǎng)絡(luò)構(gòu)建的青年女性臂部體型識(shí)別模型錯(cuò)判數(shù)分別是25、12、8個(gè)樣本,識(shí)別準(zhǔn)確率分別是85.80%、93.18%、95.45%,得出MEA-BP神經(jīng)網(wǎng)絡(luò)的識(shí)別準(zhǔn)確率最高。MEA優(yōu)化BP神經(jīng)網(wǎng)絡(luò)的權(quán)值和閾值,促進(jìn)了整體網(wǎng)絡(luò)的學(xué)習(xí)過(guò)程,提高了模型的識(shí)別精度,體現(xiàn)了思維優(yōu)化算法在優(yōu)化BP神經(jīng)網(wǎng)絡(luò)模型上的優(yōu)越性。
4結(jié)論
人體臂部體型研究是提高衣袖合體性和舒適性的關(guān)鍵,本文通過(guò)對(duì)青年女性臂部體型的研究,可以得到以下結(jié)論。
1) 應(yīng)用主成分因子分析法得到影響青年女性臂部體型的5個(gè)主成分因子,分別是臂部圍度因子、臂根因子、臂根形態(tài)因子、臂部長(zhǎng)度因子、臂部比例因子。
2) 應(yīng)用兩步聚類法,將青年女性臂部體型分為4類:長(zhǎng)胖臂、中間臂、長(zhǎng)瘦臂、短扁臂,分別占17.4%、30.7%、22.4%、29.5%。
3) 構(gòu)建了基于MEA-BP神經(jīng)網(wǎng)絡(luò)青年女性臂部體型識(shí)別模型,識(shí)別準(zhǔn)確率為95.45%,均高于單一BP神經(jīng)網(wǎng)絡(luò)和GA-BP神經(jīng)網(wǎng)絡(luò)模型。
本文將思維進(jìn)化算法(MEA)引入BP神經(jīng)網(wǎng)絡(luò)體型識(shí)別模型,不僅可以提高模型識(shí)別精度,也為采用神經(jīng)網(wǎng)絡(luò)進(jìn)行人體體型識(shí)別提供了一種新的研究方法,拓寬了MEA-BP神經(jīng)網(wǎng)絡(luò)的應(yīng)用領(lǐng)域,具有廣泛的研究?jī)r(jià)值。
參考文獻(xiàn):
[1]賀新, 金紅淑. 青年女性側(cè)面上肢形態(tài)特征細(xì)分[J]. 北京服裝學(xué)院學(xué)報(bào)(自然科學(xué)版), 2021, 41(1): 35-42.HE Xin, JIN Hongshu. Morphological feature subdivision of lateral upper limb for young female[J]. Journal of Beijing Institute of Fashion Technology (Natural Science Edition), 2021, 41(1): 35-42.
[2]張雪云. 湖北地區(qū)青年女性肩臂部特征及衣袖版型自動(dòng)生成系統(tǒng)研究[D]. 武漢: 武漢紡織大學(xué), 2017: 39-49.ZHANG Xueyun. Research on Automatic Generation of Young Women in Hubei Area and Characteristics of Shoulder and Arm Sleeve Prototype Board[D]. Wuhan: Wuhan Textile University, 2017: 39-49.
[3]舒?zhèn)? 老年女性臂部形態(tài)分析與號(hào)型規(guī)格設(shè)置[J]. 浙江紡織服裝職業(yè)技術(shù)學(xué)院學(xué)報(bào), 2021(4): 12-17.SHU Weiwei. Arm morphology analysis and type specification setting for elderly women[J]. Journal of Fashion Institute of Technology, 2021(4): 12-17.
[4]劉國(guó)聯(lián), 朱曉培, 姜川. 基于女體臂部特征的衣袖原型結(jié)構(gòu)設(shè)計(jì)研究[J]. 針織工業(yè), 2011(3): 53-55.LIU Guolian, ZHU Xiaopei, JIANG Chuan. Research on prototype structure design of sleeves based on the features of womens arms[J]. Knitting Industry, 2011(3): 53-55.
[5]YANG R Y, CHEN Y F, CUI M H. Hunchback body type classification[C]//The 14th Textile Bioengineering and Informatics Symposium Proceedings (TBIS 2021). Lieberetz: Czech Republic, 2021: 582-588.
[6]SUNG H A, LEE Y, YONG M K, et al. Fuzzy classification for Korean females lower body based on anthropometric measurement[J]. , 2016, 35(1): 121-126.
[7]劉曉航, 劉莉. BP神經(jīng)網(wǎng)絡(luò)在服裝人體工效學(xué)中的應(yīng)用[J]. 紡織導(dǎo)報(bào), 2016(9): 114-115.LIU Xiaohang, LIU Li. The application of BP neural network in clothing ergonomics[J]. China Textile Leader, 2016(9): 114-115.
[8]KONG X F, WANG X Q, GU G H, et al. Human recognition based on head-shoulder contour extraction and BP neural network[C]//Image Processing and Pattern Recognition. Beijing: International Symposium on Optoelectronic Technology and Application (IPTA), 2014.
[9]XUE S S. Research of human appearance level quantification algorithm based on BP neural network[J]. IOP Conference Series: Materials Science and Engineering, 2020, 768(7): 072021.
[10]ZHANG R Y, LEE E J. Human face recognition used improved back-propagation (BP) neural network[J]. Journal of Korea Multimedia Society, 2018, 21(4): 471-477.
[11]李康順, 李凱, 張文生. 一種基于改進(jìn)BP神經(jīng)網(wǎng)絡(luò)的PCA人臉識(shí)別算法[J]. 計(jì)算機(jī)應(yīng)用與軟件, 2014, 31(1): 158-161.LI Kangshun, LI Kai, ZHANG Wensheng. PCA face recognition algorithm based on improved BP neural network[J]. Computer Applications and Software, 2014, 31(1): 158-161.
[12]呂棟, 歐吉坤, 于勝文. 基于MEA-BP神經(jīng)網(wǎng)絡(luò)的衛(wèi)星鐘差預(yù)報(bào)[J]. 測(cè)繪學(xué)報(bào), 2020, 49(8): 993-1003.L? Dong, OU Jikun, YU Shengwen. Prediction of the satellite clock bias based on MEA-BP neural network[J]. Acta Geodaetica et Cartographica Sinica, 2020, 49(8): 993-1003.
[13]林佳能, 程仲漢, 林筆星. 基于MEA-BP的竊電識(shí)別方法研究[J]. 電子設(shè)計(jì)工程, 2021, 29(11): 175-180.LIN Jianeng, CHENG Zhonghan, LIN Bixing. Study on identification method of stolen electricity based on MEA-BP[J]. Electronic Design Engineering, 2021, 29(11): 175-180.
[14]王海軍. MEA-BP模型在遙感影像分類中的應(yīng)用研究[J]. 數(shù)學(xué)的實(shí)踐與認(rèn)識(shí), 2017, 47(2): 142-147.WANG Haijun. Research on the application of MEA-BP model in remote sensing image classification[J]. Mathematics in Practice and Theory, 2017, 47(2): 142-147.
[15]周婧, 白云龍, 張小寶, 等. 基于MEA-BP神經(jīng)網(wǎng)絡(luò)的土壤養(yǎng)分評(píng)價(jià)模型[J]. 中國(guó)農(nóng)機(jī)化學(xué)報(bào), 2020, 41(2): 231-236.ZHOU Jing, BAI Yunlong, ZHANG Xiaobao, et al. Soil nutrient evaluation model based on MEA-BP neural network[J]. Journal of Chinese Agricultural Mechanization, 2020, 41(2): 231-236.
[16]孫向東, 劉擁軍, 陳雯雯, 等. 箱線圖法在動(dòng)物衛(wèi)生數(shù)據(jù)異常值檢驗(yàn)中的運(yùn)用[J]. 中國(guó)動(dòng)物檢疫, 2010, 27(7): 66-68.SUN Xiangdong, LIU Yongjun, CHEN Wenwen, et al. Application of box plot method in outlier test of animal health data[J]. China Animal Quarantine, 2010, 27(7): 66-68.
[17]李昕, 張明明. SPSS22.0統(tǒng)計(jì)分析/從入門到精通[M]. 北京: 電子工業(yè)出版社, 2015: 273-279.LI Xin, ZHANG Mingming. SPSS22.0 Statistical Analysis/from Entry to Mastery[M]. Beijing: Publishing House of Electronics Industry, 2015: 273-279.
[18]王軍, 李曉久, 潘力, 等. 東北地區(qū)青年女性腰臀部體型特征與分類[J]. 紡織學(xué)報(bào), 2018, 39(4): 106-110.WANG Jun, LI Xiaojiu, PAN Li, et al. Waist hip somatotype and classification of young women in northeast China[J]. Journal of Textile Research, 2018, 39(4): 106-110.
[19]王小川, 史峰, 郁磊, 等. MATLAB神經(jīng)網(wǎng)絡(luò)43個(gè)案例分析[M]. 北京: 北京航空航天大學(xué)出版社, 2013: 265-278.WANG Xiaochuan, SHI Feng, YU Lei, et al. MATLAB Neural Network Analysis of 43 Cases[M]. Beijing: Beihang University Press, 2013: 265-278.
[20]SUN C Y, SUN Y, XIE K M. Mind-evolution-based machine learning and applications[C]//IEEE. Hefei: Proceeding of the 3rd World Congress on Intelligent Control and Automation, 2000: 112-117.
[21]WANG Z H, MA G S, GONG D Y, et al. Application of mind evolutionary algorithm and artificial neural networks for prediction of profile and flatness in hot strip rolling process[J]. Neural Processing Letters, 2019, 50(3): 2455-2479.
[22]周振民, 劉荻. 基于Matlab的人工神經(jīng)網(wǎng)絡(luò)用水量預(yù)測(cè)模型[J]. 中國(guó)農(nóng)村水利水電, 2007(4): 45-47.ZHOU Zhenmin, LIU Di. Artificial neural network water consumption prediction model based on Matlab[J]. China Rural Water and Hydropower, 2007(4): 45-47.
[23]孫俊, 唐凱, 毛罕平, 等. 基于MEA-BP神經(jīng)網(wǎng)絡(luò)的大米水分含量高光譜技術(shù)檢測(cè)[J]. 食品科學(xué), 2017, 38(10): 272-276.SUN Jun, TANG Kai, MAO Hanping, et al. Hyperspectral detection of moisture content in rice based on MEA-BP neural network[J]. Food Science, 2017, 38(10): 272-276.
Young female arm body shape recognition based on MEA-BP neural network
NI Shiming BAI Yunlong JIANG Yiqun
(1a.College of Architecture & Art Design; 1b.College of Information Engineering, Shaoxing Vocational & Technical College,
Shaoxing 312000, China;
2.College of Engineering, Iowa State University, Iowa 50011, USA)
Abstract: The research on human arm shape is the key to improving the fit and comfort of the sleeve. Young females arm shape has multiple nonlinear features. Since the initial threshold and weight are randomly selected, the single BP neural network is prone to the problems of unstable model and low accuracy when dealing with the complex nonlinear relationship of human body shape. The thought evolution algorithm is a global optimization algorithm, which can optimize the initial threshold and weight of BP neural network through multiple "convergence" and "alienation" operations, so as to improve the stability of the model and prediction accuracy. In order to quickly and accurately identify a young female arm shape, this paper constructs a MEA optimized BP neural network based arm shape recognition model for young females.
Firstly, the arm data of 611 young women aged 18-25 are obtained by [TC]2 3D body measurement, and the values of 14 measurement items are extracted. SPSS software is used to preprocess the data, and descriptive statistical analysis of the indicators is conducted. Secondly, through principal component factor analysis, five morphological factors affecting young female arm shape are obtained, namely, arm circumference factor, arm root factor, arm root morphology factor, arm length factor and arm proportion factor. The characteristic factors corresponding to the maximum load of the five factors are: upper arm circumference, root circumference, root flattening rate, whole arm length and upper arm length/forearm length, which are selected and divided into five categories by two-step clustering method: long fat arm, middle arm, long thin arm and short flat arm, accounting for 17.4%, 30.7%, 22.4% and 29.5%, respectively. Finally, Matlab software is used to construct a young female arm body shape recognition model based on MEA-BP neural network. MEA optimizes the weight and threshold value of BP neural network, which promotes the learning process of the whole network and improved the stability and recognition accuracy of the model. The innovation of this paper is reflected in the research object and research method. Existing studies are based on the subdivision and cross combination of typical indicators to obtain the complex arm shape, but there are a large number of body shape subdivisions, and there is a lack of studies on the overall classification and recognition of the arm shape. In this paper, it is an innovation to select the arm shape for the whole classification and recognition research. The MEA-BP neural network has a good application in satellite clock prediction, electric stealing identification, remote sensing image classification, soil nutrient grading evaluation and other fields. This paper tries to introduce the thought evolution algorithm (MEA) optimized BP neural network into the field of human engineering for the identification of young female arm shapes, which is an innovation of research method. Compared with the single BP neural network and GA-BP neural network, the MEA-BP neural network model has higher prediction accuracy and better nonlinear mapping ability.
The introduction of mind evolution algorithm (MEA) into BP neural network body shape recognition model can not only improve the accuracy of the model recognition, but also provide a new research method for body shape recognition using neural network, which broadens the application field of MEA-BP neural network, and has a wide range of research value.
Key words: young females; arm shape; body type classification; MEA-BP neural network; recognition model