基于Tensorflow框架的手寫數(shù)字識別
打開文本圖片集
摘要:文章利用Tensorflow深度學(xué)習(xí)結(jié)構(gòu)來構(gòu)建神經(jīng)網(wǎng)絡(luò)模型,并采用激活函數(shù)對MINIST進(jìn)行訓(xùn)練;加入特征轉(zhuǎn)換過程,利用梯度下降優(yōu)化器,將數(shù)據(jù)降維;在輸出層上將全連接模型和Softmax層相結(jié)合,經(jīng)過交叉驗(yàn)證,達(dá)到90%以上的識別率。
關(guān)鍵詞:Tensorflow;MNIST;梯度下降優(yōu)化器;全連接模型
doi:10.3969/J.ISSN.1672-7274.2023.02.044
中圖分類號:TP 3 文獻(xiàn)標(biāo)示碼:A 文章編碼:1672-7274(2023)02-0-04
Handwritten Digit Recognition Based on Tensorflow Framework
LI Linfeng, CHEN Jiayi, ZHENG Jiawei, LI Tong, WU Junqin
(School of Computer Science, Southwest Petroleum University, Chengdu 610500, China)
Abstract: This paper uses Tensorflow deep learning structure to build neural network model, and uses activation function to train MINIST; Add feature transformation process and use gradient descent optimizer to reduce data dimension; In the output layer, the full connection model is combined with the Softmax layer. After cross validation, the recognition rate is more than 90%.
Key words: Tensorflow; MINIST; gradient descent optimizer; full connection layer
0 引言
手寫體數(shù)字識別的實(shí)質(zhì)就是預(yù)測,預(yù)測出該圖片屬于哪一個(gè)類別就認(rèn)為這幅圖片則為哪一個(gè)數(shù)字。(剩余4647字)