基于BP神經(jīng)網(wǎng)絡的手寫數(shù)字識別(完整版)doc資料

于神經(jīng)網(wǎng)絡的寫數(shù)字識別（完整版）資料BPrDigitRecognitiBasedonNeuralNetwork2

072650;2.

071003[]

,

BP,

BP

MATLABBP[]

;BP[]TP391.41;TP183[]A,到交教育郵政給人們便算,運流以計聚析算為如Bagging差反播算持以人,為BP,

BP1BP誤反反如

1,,多BP,,,,1BP2BPBP,BP22—32—2.1

[1],,graythreshim=graythresh(fig_gray=i(fig_gray,thresholdfig_grayfig_bool3

,,

,20,720×1,matlab=fig_bool_normalizedfig_bool_normalized[2]1010當像入神經(jīng)網(wǎng)絡在經(jīng)對應位置上其他位置00,1經(jīng)其他入第2經(jīng)元其他以類。隱含層多神經(jīng)網(wǎng)學習速度就越,據(jù)Kosmog定理在合理構和恰當值條件下,網(wǎng)絡可逼近任意續(xù)函我們結構相簡層BP網(wǎng)絡。一般情況,層神經(jīng)元是根據(jù)絡收斂性能壞來確定在結大量絡構礎上得經(jīng)公式s=0.43nm+0.12m其中層經(jīng)元,m神經(jīng)元,據(jù)以上公可得隱含層神經(jīng)元建立一向神經(jīng)網(wǎng)絡函newff=newff[720,70,10],{‘logsig’,(‘logsig‘logsig’},‘traincgb’;minmax720P[720,70,10]{ogsig’,′traincgb′,Powell2Beale2.4

,36×20=720,0~910720×10,:sampoup,91,matlab:targets=eye(10;10,2020:P=[sales_group1,sales_group2,42=double(P,,

20SSE,0.01,0.01,。序代如[4-5]:.trainPara=1000;%.trainPara顯示隔.trainParam.min_grad-10;%小行度net.perfor=‘ssenet.trainParam.goalnet.layers{1}.initFcn=‘initb層初始化‘initwb’,使面輸入層始語句‘randnr’有效net.inputeights{1,1}.initFcn=‘randnr輸入層權向量始化net.layereights{2,1}.initFcn=‘randnr第1層2向量初始化net(net;%始其中[net,tr]更新權,tr記(果如圖4所4

,SSE0.00882.5

sit=si(net,test_salemp,testtle=dou;%outm(net,test_sa;%_ve(testt;%if(max_valu>0.7%0.7,testout=competou;numberouend3

1020200,99.48%,,95.5%,0.71

020210100%15%95%120210100%15%95%220210100%00100%320210100%315%85%420210100%15%95%520210100%15%95%620210100%00100%720210100%00199.48%84%30—523

ABSECU,,,ABS4,ABS

4ABS30k/h,4ABS,0,,ABS4ABS4

ABSABSABS,,ABSABS,[1].ABS[],2020(2:77279.[2]鄭太雄李銳蔣ABS單元[].慶學報(自然版,[3]

.ABS].,2001(5:15217.[4]

.(ABS[M].,1996.[5]ABS[M].:,2002:1052107.[7]

.LabV[J

].,2003,18(2:9211.[10],,.LabVIE[M].:254

BP