Multi stream CNN implementation

2 次查看（过去 30 天）
Rd
Rd 2020-11-19
评论： Ankita Jain 2022-3-3
i have designed CNN and it is trained with dataset stored in XTrain image datastore.
layers = [
imageInputLayer([227 227 3],"Name","imageinput")
.
.
.
averagePooling2dLayer([5 5],"Name","avgpool2d_5","Padding","same","Stride",[2 2])
fullyConnectedLayer(10,"Name","fc")
softmaxLayer("Name","softmax")
classificationLayer("Name","classoutput")];
options = trainingOptions('sgdm', ...
'LearnRateSchedule','piecewise', ...
'InitialLearnRate',0.01, ...
'ValidationFrequency',10,...
'MaxEpochs',10, ...
'MiniBatchSize',10, ...
'ValidationData',{input,testinput},...
'Plots','training-progress')
[net, traininfo] = trainNetwork(input,layers,options);
YPred = classify(net,testinput);
accuracy = mean(YPred==YTest)
confusion = plotconfusion(YTest,YPred)
I got a good result.
I have to train my CNN with three different datasets namely input_in, input_mid, input_rin. Reference block is attached below.
After last pooling layer i have to concatenate the features. Then fully connect, softmax and classification layer should appear.
i have try like this:
deepnet1=Layer(input_in);
deepnet2=Layer(input_mid);
deepnet3=Layer(input_rin);
deepnet4=createLayerFullyConnect(numHiddenDimension);
function deepnet=createLayerFullyConnect(numHiddenDimension)
layers = [
imageInputLayer([1 numHiddenDimension*3 1],"Name","imageinput","Normalization","none")
fullyConnectedLayer(30,"Name","fc_1")
fullyConnectedLayer(20,"Name","fc_2")
fullyConnectedLayer(10,"Name","fc_3")];
lgraph = layerGraph(layers);
deepnet = dlnetwork(lgraph);
end
I don't know how to train same network with different images and concatenate them after training. Any help is appriciated. Thanks in advnace.
Regards,
Ramasenthil.

请先登录，再回答此问题。

回答（0 个）

类别

在 Help Center 和 File Exchange 中查找有关 Recognition, Object Detection, and Semantic Segmentation 的更多信息

标签

Community Treasure Hunt

Find the treasures in MATLAB Central and discover how the community can help you!

Start Hunting!

Translated by