Convolutional Neural Network (CNN) & nws yuav siv hauv nab hab sej

CNN thiab Python – Yuav ua li cas rau nws?

Taw qhia :

Cov hnub Convolutional neural tes hauj lwm (CNNs/ConvNets) Yog ib lub npe kub rau lub computer kev tshawb fawb uas tseem nyob rau hauv lub minds ntawm lub lag luam thiab nws cov thawj coj tshwj xeeb tshaj yog rau nws siv nyob rau hauv kev tsim lag luam thiab raws li kev siv cov ntaub ntawv scientific. Nws yuav pom raws li lub tshuab kawm txheej txheem los yog ib tug sib sib zog nqus kev kawm. Yog koj tsis paub txog lub tshuab kawm (ML) thiab sib sib zog nqus (DL) Ces tsis txhob txhawj yuav muaj ib nyuag ntsia ntawm ob leeg, ntxiv, nrog CNN & nws yuav siv hauv nab hab sej. Feem ntau cov CNNs siv tau nyob rau hauv cov ntaub ntawv hais txog tej yam kev hais lus, Duab thiab raws li yees duab recognition. Nyob rau cov tshooj no, Peb yuav tshawb CNN nrog rau lub architecture qab nws thiab ib qho piv txwv, Hais txog cov duab classifier, DVR hauv Python, Piv txwv tias koj muaj yooj yim to taub txog qhov chaw ua hauj lwm ntawm neural tes hauj lwm. Peb yuav zoo rau nws taug kev los ntawm cov lus no kom koj to taub cov hauj lwm ntawm CNNs tsaug. Wb pib peb cov lus:

Evolution of Machine Learning (ML) & Kev kawm sib sib zog nqus (DL)

Ua ntej digging lub tswvyim ntawm CNNs, Peb yuav tsum muaj ib tug ceev saib rau lub evolution ntawm ML & DL. kev kawm tshuab (ML), daim ntawv thov artificial txoj kev ntse, lawm ib unheard npe, Txawm li cas los, Nws muaj zoo kawg li evolved hla lub xyoo caum lub xeem. Npaum daim credit rau tib mus rau qhov kev kho tshiab ntawm hardware thiab algorithms. Lub evolution ntawm ML pib ntawm cov qauv recognition uas muab lub computer systems muaj peev xwm kawm los ntawm cov ntaub ntawv thiab txhim kho los ntawm kev tsis tau explicitly programmed ua tej hauj lwm zoo.

Rau lwm cov tes, Lub evolution ntawm kev sib sib zog nqus pib ntawm cov kev kho tshiab hauv neural tes hauj lwm. Raws li yog ib hom qauv rau tshuab kawm, mus-1980s & Thaum ntxov Kwvyees li xyoo 1990 neural tes hauj lwm tau ncaim ntawm tej explosive computational kev loj hlob thiab raws li architectural advancements. Yog vim li cas deep kawm, emerged ntawm ntawd computational kev loj hlob, yog ib hom cov kev kawm tshuab Nws algorithms yog inspired los ntawm cov functionality ntawm tib neeg lub hlwb.

Koj kuj tseem nyiam nyeem ntawv – Kev kawm nyob rau hauv cov kev kawm tshuab

Xyuas peb lub tshuab kawm archive – Tshuab kawm archive

Xyuas peb artificial txoj kev ntse archive – AI archive

Elucidation of Convolutional Neural Networks (CNNS)

Convolutional neural tes hauj lwm (CNNS) Zoo tib yam neural tes hauj lwm (NNS) Nyob rau hauv tus txiv neej uas lawv tseem ua los ntawm neurons uas muaj learnable nyhav thiab biases. Recalling qhov chaw ua hauj lwm ntawm neural tes hauj lwm, Txhua neuron hauv lub network tau txais ib los yog ntau inputs, Yuav siv sij hawm ib weighted summ uas yog ntxiv kis los ntawm ib qho kev ua hauj lwm activation los ua cov zis kawg nkaus. Yog CNNs thiab zoo tib yam NNs muaj coob zoo sib xws ces cov nqe lus nug arises dab tsi yeej ua rau lawv sib txawv?

Cov ntaub ntawv kho cov ntaub ntawv thiab hom khaubncaws sab nraud ua rau lawv sib txawv. Ordinary neural tes hauj lwm las mees cov qauv ntawm cov ntaub ntawv input thiab tag nrho cov ntaub ntawv yog converted rau hauv 1-D array ua ntej pub nws mus rau hauv lub network. Rau lwm cov tes, CNN tus architecture yog tsim nyob rau hauv tej ib txoj kev uas nws yuav coj mus ua tus account 2D qauv ntawm cov dluab (los yog lwm yam 2D input xws li hais lus) Thaum processing lawv thiab kuj pub nws extract cov khoom rau dluab. Ntxiv, Yog peb tham txog cov khaubncaws sab nraud ces CNNs muaj qhov kom zoo dua uas muaj ib los yog ntau convolutional layers thiab pas dej ua ke txheej txheem (Lub tsev loj blocks ntawm CNNs) Raws li ib tug los yog ntau siab txuas nrog txheej txheem li txheem multilayer neural tes hauj lwm. Nws txhais tau tias peb yuav xav txog CNN ua ib rooj plaub tshwj xeeb uas siab txuas nrog tes hauj lwm. nthuav, Tsis yog nws?

Architecture ntawm Convolutional Neural Networks (CNNS)

CNN architecture no tsuas yog ib daim ntawv teev cov khaubncaws sab nraud transforming lub 3D (width, qhov siab thiab qhov tob) Duab ntim rau hauv ib lub tso zis 3D tso zis ntim. Txhua neuron nyob rau hauv cov txheej txheem tam sim no yog txuas nrog ib thaj me me ntawm cov zis los ntawm cov txheej txheem yav dhau los. Nws zoo li overlaying ib filter rau cov duab input. Nws siv M filters kom paub tseeb txog tag nrho cov lus. Cov M filters no yeej feature extractors uas extract nta xws li edges, ces kaum thiab li ntawd. Ua ntej yuav sib sib zog nqus architecture, Kuv xav piav cov khaubncaws sab nraud [INPUT-CONV-RELU-POOL-FC] Uas siv los tsim CNNs:

INPUT: Txheej txheem no, Raws li lub npe implies, Yuav tuav tus nyoos pixel qhov tseem ceeb i.e.e.. Cov ntaub ntawv ntawm cov duab raws li nws yog. Piv txwv, INPUT [64×64×3] Txhais tau tias cov duab RGB ntawm width 64, qhov siab 64 qhov tob 3. Nws yog ib cov duab 3-channeled duab.
CONV: Feem ntau cov computation, convolutions ntawm neurons thiab ntau patches hauv lub input, Ua li cas nyob rau hauv Convolutional txheej txheem li no nws yog ib lub tsev blocks ntawm CNNs. Piv txwv, Yog peb txiav txim siab siv 6 filters rau saum toj no hais input ces qhov no tej zaum yuav ua rau lub ntim [[64×64×6].
RELU: Nws hu ua rectified linear chav tsev uas yeej siv ib qho kev ua hauj lwm activation rau cov zis yav dhau los. Ib tug tsis-linearity yuav muab ntxiv rau lub network los RELU.
PAS DEJ UA KE: Pas dej ua ke yog lwm lub tsev thaiv ntawm CNN. Nws lub ntsiab hauj lwm tseem ceeb yog down-sampling uas nws koomtxoos ywj siab rau txhua slice ntawm cov input thiab resizes nws spatially.
FC: Siab txuas nrog txheej txheem los yog ntau specifically hu ua tso zis txheej txheem siv los laij cov hoob kawm score. Cov zis tso zis yog ntim ntawm qhov luaj li cas uas L yog tus xov tooj corresponding rau hoob kawm tau.

Tom qab diagram nruab nrab yog tus raug architecture ntawm CNNs:

Duab 1: Architecture of CNNs

Piv txwv: Ib cov duab classifier DVR hauv Python

Os, Peb yuav siv cov duab classifier siv CNN hauv Python. Lub tswvyim yooj yim yuav zoo li ces nws yuav tau thov kom muaj ntaub ntawv xws li ntuj hom lus (NLP), Yees duab recognition thiab lwm yam kev siv rooj plaub ib yam nkaus thiab. Siv qhov kev siv rau qhov no, Peb muaj cov nram qab no prerequisites:

keras: Ib lub tsev qiv ntawv sib sib zog nqus tsev kawm ntawv: – Nws yog ib theem neural network API uas yog sau nyob rau hauv Python thiab muaj peev xwm khiav saum toj kawg nkaus TensorFlow, CNTK los yog Theno. Yog hais tias koj xav kawm nws kom paub meej ces mus rau qhov txuas https://keras.io/. Nruab nrab ntawm keras kev sib sib zog nqus tsev kawm ntawv yuav tsum xub pib siv nws. Tom qab commands yuav siv tau los nruab nrab nws:

ntsia keras

Rau conda ib puag ncig command yuav ua raws li:

nqi sib ntxiv –c-forge Keras

Kev kawm & Kuaj cov ntaub ntawv teev cov dluab: – Peb kuj xav tau kev cob qhia & Kuaj cov ntaub ntawv teev cov dluab. Kuv siv cov dluab ntawm miv thiab dev los ntawm qhov txuas https://www.kaggle.com/c/dogs-vs-cats/data.

Tam sim no, Tom qab no Python tsab ntawv los siv cov duab classifier:

Qhia 1: Python tsab ntawv los siv cov duab classifier

[chaws]

Ntawm keras.qauv import Sequential

Ntawm keras.layers import Conv2D

Ntawm keras.layers import MaxPooling2D

Ntawm keras.layers import Flatten

Ntawm keras.layers import Dense

Img_classifier = Sequential()

Img_classifier ntxiv(Conv2D(32, (3, 3), input_shape = (64, 64, 3, activation = 'relu'))

Img_classifier ntxiv(MaxPooling2D(pool_size = (2, 2)))

Img_classifier ntxiv(flatten())

S_classifier.add(dense(chav = 128, activation = 'relu'))

Img_classifier ntxiv(dense(chav = 1, activation = 'sigmoid'))

Img_classifier.compile(optimizer = 'adam', poob = 'binary_crossentropy', metrics = ['accuracy'])

training_datagen = ImageDataGenerator(ua tsis taus pa = 1./255,shear_range = 0.2, zoom_range = 0.2,horizontal_flip = True)

testing_datagen = ImageDataGenerator(rov qab = 1./255)

training_set = training_datagen.flow_by_directory("/Cov neeg siv/admin/training_set",target_size = (64, 64),batch_size = 32,class_mode = 'binary')

test_set = testing_datagen.flow_by_directory('test_set',target_size = (64, 64),batch_size =32,class_mode = 'binary')

classifier.fit_generator(training_set,steps_per_epoch = 8000,epochs = 25,validation_data = test_set,validation_steps = 2000)

Ntawm keras.preprocessing import duab

test_image = image.load_img('dataset/single_prediction/cat_or_dog_1.jpg', target_size = (64, 64))

test_image = image.img_to_array(test_image)

test_image = np.expand_dims(test_image, axis = 0)

= classifier.predict(test_image)

training_set.class_indices

yog tshwm sim[0][0] == 1:

twv ua ntej = 'aub’

ntxiv:

twv ua ntej = 'miv’

[/chaws]

Xaus tsab ntawv

Kuv vam tias koj muaj enjoyed no luv lus ntawm kev kawm ua hauj lwm ntawm CNN nrog kuv. Sim mus cuag cov lus no los ntawm lub tsev koj tus kheej CNN network thiab thov tib yam rau ntau daim ntawv thov ib yam nkaus thiab. Yog hais tias koj tuaj hla tej teeb meem nyuaj thaum uas siv CNNs hauv Python, los yog koj muaj tej tswv yim / feedback thov xav dawb rau ncej lawv nyob rau hauv cov nqe lus hauv qab no. Kuv yuav nyiam hnov ntawm koj.

Tej zaum koj kuj nyiam nyeem cov dab neeg nram qab no.

Sau Bio: Gaurav Leekha Yog ib passionate technical ntsiab lus txawj sau ntawv, Feem ntau rau teb ntawm txoj kev ntse Artificial txoj kev ntse (AI), Cov kev kawm tshuab (ML), Kev kawm sib sib zog nqus, Artificial Neural Networks (IB XYOOS TWG), Hais lus, thiab nab hab sej. Nws muaj 7+ Xyoo kev qhia ntawv nyob rau hauv daim teb ntawm computer kev kawm. Nws tseem pursuing Ph.D. Kawm ntawv nyob rau hauv daim teb ntawm tshuab kawm thiab nrog rau cov kev ntsuam xyuas ntawm ntau lub teb chaws thiab raws li thoob ntiaj teb journals xws li International Journal of hais lus tshuab (KUV XYAUM), caij nplooj ntoos hlav.