《電子技術(shù)應(yīng)用》
您所在的位置:首頁 > 人工智能 > 設(shè)計(jì)應(yīng)用 > 基于改進(jìn)LeNet-5的形狀類似物體識(shí)別方法
基于改進(jìn)LeNet-5的形狀類似物體識(shí)別方法
《信息技術(shù)與網(wǎng)絡(luò)安全》2020年第6期
鄭 睿1,2,,余 童1,,2,程龍閱1
1.安徽師范大學(xué) 物理與電子信息學(xué)院,安徽 蕪湖241002; 2.安徽省智能機(jī)器人信息融合與控制工程實(shí)驗(yàn)室,安徽 蕪湖241000
摘要: 針對(duì)深度學(xué)習(xí)在對(duì)外形類似物體的識(shí)別上存在著識(shí)別精度低,、耗時(shí)長(zhǎng)等問題,提出基于改進(jìn)的LeNet-5的識(shí)別方法,。在傳統(tǒng)LeNet-5網(wǎng)絡(luò)基礎(chǔ)上,,將卷積層變?yōu)殡p層非對(duì)稱卷積使網(wǎng)絡(luò)有更好的特征提取能力;通過批量歸一化提高網(wǎng)絡(luò)泛化能力,;采用全局平均池化替代原Flatten層,,用于克服傳統(tǒng)全連接層參數(shù)多、耗時(shí)長(zhǎng)的缺點(diǎn),;通過對(duì)訓(xùn)練集進(jìn)行增廣增加訓(xùn)練樣本,。實(shí)驗(yàn)結(jié)果表明,改進(jìn)LeNet-5網(wǎng)絡(luò)的訓(xùn)練精度達(dá)到91%,,識(shí)別形狀類似物體的精度為87%,,且能在較少迭代次數(shù)內(nèi)收斂,這些指標(biāo)均顯著優(yōu)于原網(wǎng)絡(luò),。
中圖分類號(hào): TP183
文獻(xiàn)標(biāo)識(shí)碼: A
DOI: 10.19358/j.issn.2096-5133.2020.06.006
引用格式: 鄭睿,,余童,程龍閱. 基于改進(jìn)LeNet-5的形狀類似物體識(shí)別方法[J].信息技術(shù)與網(wǎng)絡(luò)安全,,2020,,39(6):31-37,,43.
Recognition method of similar-shaped objects based on improved LeNet-5
Zheng Rui1,2,,Yu Tong1,,2,Cheng Longyue1
1.College of Physics and Electronic Information,,Anhui Normal University,,Wuhu 241002,China,; 2.Anhui Province Engineering Laboratory of Intelligent Robot′s Information Fusion and Control,Wuhu 241000,,China
Abstract: Aiming at the problems of low recognition accuracy and long time-consuming in the recognition of similar shape objects by deep learning, a recognition method based on improved LeNet-5 is proposed. Based on the traditional LeNet-5 network, changing the convolutional layer into a double-layer asymmetric convolution makes the network have better feature extraction capabilities; the generalization ability of the network is improved by batch normalization; the original Flatten layer is replaced by global average pooling,,which is used to overcome the shortcomings of the traditional fully-connected layer with many parameters and long time-consuming; the training sample is increased by augmenting the training set. Experimental results show that the training accuracy of the improved LeNet-5 network reaches 91%, the accuracy of identifying objects with similar shapes is 87%, and it can converge within a small number of iterations. These indicators are significantly better than the original network.
Key words : LeNet-5 network;image recognition,;asymmetric convolution,;batch normalization;maximum average pooling

當(dāng)前,,基于視覺的智能機(jī)器人已經(jīng)應(yīng)用于各領(lǐng)域中,。當(dāng)機(jī)器人面臨需要抓取形狀類似、硬度不同的物體時(shí),,智能機(jī)器人應(yīng)選擇不同的抓取力,。因此,通過視覺識(shí)別出這類物體具有較高實(shí)用價(jià)值,。利用深度學(xué)習(xí)對(duì)圖像進(jìn)行識(shí)別是較為高效的方法,,國(guó)內(nèi)外相關(guān)研究已經(jīng)在車輛及車道線檢測(cè)、人臉識(shí)別,、手寫體識(shí)別等領(lǐng)域取得較多的成果,。

近年來,針對(duì)形狀類似物體的識(shí)別也取得了一定的成果,。張雪芹等人利用深度學(xué)習(xí)AlexNet網(wǎng)絡(luò)實(shí)現(xiàn)了對(duì)多種類植物圖片進(jìn)行分類識(shí)別,。林思思等提出融合深度特征和人工特征的花卉圖像特征提取方法,并在此基礎(chǔ)上實(shí)現(xiàn)花卉圖像的分類,。西南交通大學(xué)秦放提出基于深度學(xué)習(xí)的昆蟲圖像識(shí)別研究,,擴(kuò)充了昆蟲樣本集,基于昆蟲圖像識(shí)別的任務(wù)需求和樣本集,,從網(wǎng)絡(luò)和訓(xùn)練兩個(gè)方面進(jìn)行改進(jìn),。張立超等人利用LeNet-5網(wǎng)絡(luò)對(duì)兩種品種的蘋果進(jìn)行分類識(shí)別,在兩種蘋果的分類中取得不錯(cuò)的效果,。但大型的神經(jīng)網(wǎng)絡(luò)一般通過加深網(wǎng)絡(luò)深度獲得高精度的識(shí)別率,,其結(jié)構(gòu)較為復(fù)雜,,運(yùn)算量大,無法滿足快速識(shí)別的要求,。

因此,,本文選擇LeNet-5網(wǎng)絡(luò)作為研究對(duì)象,由于LeNet-5網(wǎng)絡(luò)結(jié)構(gòu)較為簡(jiǎn)單,,運(yùn)算量較小,,對(duì)硬件配置要求低,能夠在滿足網(wǎng)絡(luò)輕量化的前提下對(duì)形狀類似,、硬度不同的物體實(shí)現(xiàn)快速識(shí)別,。傳統(tǒng)LeNet-5網(wǎng)絡(luò)在識(shí)別精度上尚有不足。為了能夠提高識(shí)別精度且盡可能地減少運(yùn)算量,,本文將傳統(tǒng)卷積核拆分為非對(duì)稱卷積核以縮短計(jì)算時(shí)間,;在網(wǎng)絡(luò)中間加入BN層使中間層的數(shù)據(jù)輸出更加穩(wěn)定,提高網(wǎng)絡(luò)的訓(xùn)練精度,;采用全局平均池的方法取代原模型Flatten層以降低運(yùn)算量,。通過這三種方式有效地改進(jìn)了LeNet-5網(wǎng)絡(luò),使其能夠適用于移動(dòng)機(jī)器人平臺(tái),,實(shí)現(xiàn)對(duì)物體的識(shí)別,。



本文詳細(xì)內(nèi)容請(qǐng)下載:  http://forexkbc.com/resource/share/2000003185

作者信息:

鄭  睿1,2,,余  童1,,2,程龍閱1

(1.安徽師范大學(xué) 物理與電子信息學(xué)院,,安徽 蕪湖241002,;

2.安徽省智能機(jī)器人信息融合與控制工程實(shí)驗(yàn)室,安徽 蕪湖241000)


此內(nèi)容為AET網(wǎng)站原創(chuàng),,未經(jīng)授權(quán)禁止轉(zhuǎn)載,。