一種基于部件功用性語義組合的家庭日常工具分類方法
發(fā)布時(shí)間：2018-12-29 16:15
【摘要】：為滿足人機(jī)共融環(huán)境下機(jī)器智能對(duì)工具功用性認(rèn)知的需要,模擬人類自底向上的認(rèn)知方式,設(shè)計(jì)了一種基于部件功用性語義組合的聚類方法,來對(duì)家庭日常工具進(jìn)行表示與建模.首先,設(shè)計(jì)了工具功用性部件邊緣表示方法并基于結(jié)構(gòu)隨機(jī)森林加以建模.然后基于功用性部件組合思想,設(shè)計(jì)了高層語義空間上聯(lián)合各部件顯著度的工具整體表示方法并采用聚類方式構(gòu)建工具功用性字典.在線檢測(cè)階段,聯(lián)合測(cè)試樣本各功用性部件的顯著度,利用其與工具功用性字典的距離殘差對(duì)工具進(jìn)行分類判別.在實(shí)驗(yàn)中,將7種功用性部件組合聚類形成5類工具,當(dāng)各類工具選取不同核值時(shí),分類精度可達(dá)90%以上,即使各類工具的核值固定為3,分類精度也在85%以上.實(shí)驗(yàn)結(jié)果表明,相較于傳統(tǒng)的特征表示方式,功用性語義的加入使機(jī)器人深化了對(duì)工具功能的認(rèn)知,基于功用性部件組合的字典表示使得家庭常見工具的分類精度和效率明顯提升,且實(shí)現(xiàn)了工具間功能相似性測(cè)算和最優(yōu)替代工具查找.
[Abstract]:In order to meet the needs of machine intelligence for tool functional cognition in human-computer inclusive environment, a clustering method based on component functional semantic combination was designed to simulate human bottom-up cognition. To represent and model everyday family tools. Firstly, the edge representation method of utility components is designed and modeled based on structured random forest. Then, based on the idea of functional component combination, a tool overall representation method combining the saliency of each component in high-level semantic space is designed, and a tool functional dictionary is constructed by clustering method. In the online detection stage, the salience of each functional part of the sample is jointly tested, and the distance residuals between the sample and the tool function dictionary are used to classify the tool. In the experiment, the 7 kinds of functional components are grouped into five kinds of tools. When different kernel values are selected, the classification accuracy can reach more than 90%. Even if the kernel value of all kinds of tools is fixed at 3, the classification accuracy is more than 85%. The experimental results show that, compared with the traditional feature representation, the addition of functional semantics makes the robot deepen its understanding of the tool function. The dictionary representation based on functional component combination improves the classification accuracy and efficiency of common family tools, and realizes the function similarity measurement and optimal alternative tool search among tools.
【作者單位】： 燕山大學(xué)信息科學(xué)與工程學(xué)院;中國(guó)科學(xué)院自動(dòng)化研究所復(fù)雜系統(tǒng)管理與控制國(guó)家重點(diǎn)實(shí)驗(yàn)室;河北省計(jì)算機(jī)虛擬技術(shù)與系統(tǒng)集成重點(diǎn)實(shí)驗(yàn)室;
【基金】：國(guó)家自然科學(xué)基金(61305113) 河北省自然科學(xué)基金(F2016203358)
【分類號(hào)】：TP242
，
本文編號(hào)：2395050
[Abstract]:In order to meet the needs of machine intelligence for tool functional cognition in human-computer inclusive environment, a clustering method based on component functional semantic combination was designed to simulate human bottom-up cognition. To represent and model everyday family tools. Firstly, the edge representation method of utility components is designed and modeled based on structured random forest. Then, based on the idea of functional component combination, a tool overall representation method combining the saliency of each component in high-level semantic space is designed, and a tool functional dictionary is constructed by clustering method. In the online detection stage, the salience of each functional part of the sample is jointly tested, and the distance residuals between the sample and the tool function dictionary are used to classify the tool. In the experiment, the 7 kinds of functional components are grouped into five kinds of tools. When different kernel values are selected, the classification accuracy can reach more than 90%. Even if the kernel value of all kinds of tools is fixed at 3, the classification accuracy is more than 85%. The experimental results show that, compared with the traditional feature representation, the addition of functional semantics makes the robot deepen its understanding of the tool function. The dictionary representation based on functional component combination improves the classification accuracy and efficiency of common family tools, and realizes the function similarity measurement and optimal alternative tool search among tools.
【作者單位】： 燕山大學(xué)信息科學(xué)與工程學(xué)院;中國(guó)科學(xué)院自動(dòng)化研究所復(fù)雜系統(tǒng)管理與控制國(guó)家重點(diǎn)實(shí)驗(yàn)室;河北省計(jì)算機(jī)虛擬技術(shù)與系統(tǒng)集成重點(diǎn)實(shí)驗(yàn)室;
【基金】：國(guó)家自然科學(xué)基金(61305113) 河北省自然科學(xué)基金(F2016203358)
【分類號(hào)】：TP242
，
本文編號(hào)：2395050
本文鏈接：http://sikaile.net/kejilunwen/zidonghuakongzhilunwen/2395050.html
最近更新
教材專著
