MATLAB: Feature Exctraction using GoogLeNet with SVM

svm

Hi, there! I`m very new to matlab.

I run the GoogLenet code based on following link. It works well, however it take lots of time in case of big data.

reference:https://jp.mathworks.com/help/deeplearning/examples/train-deep-learning-network-to-classify-new-images.html

Therefore, I would like to extract learned image features from a pretrained convolutional neural network, and use those features to train an image classifier. And also, I have a plan to train a support vector machine (SVM) using fitcecoc (Statistics and Machine Learning Toolbox™) on the extracted features

Here is feature extraction code using Alexnet with SVM.

reference:https://jp.mathworks.com/help/deeplearning/examples/feature-extraction-using-alexnet.html?lang=en

I would like to run the feature extraction code using GoogLeNet with SVM, but not sure how to implement…

Could you have any tip to implement it?

Thank you in advance!

Best Answer

I was having the same problem.

See if this can help: https://www.mathworks.com/help/deeplearning/ref/alexnet.html

Check the chapter "Feature Extraction Using AlexNet" in that link, I think it also works with GoogleNet.

Good luck!

Related Solutions

MATLAB: How to extract features from different layers in GoogLeNet

use some layer as the feature . then you can use it for other classifier

    net = googlenet;
    inputSize = net.Layers(1).InputSize;
    imds = imageDatastore(dbpathSave)
    augimdsTrain = augmentedImageDatastore(inputSize(1:2),imds );
    %%Extract Image Features
    layer = 'loss3-classifier';%1000
    featuresTrain = activations(net,augimdsTrain,layer,'OutputAs','rows');

MATLAB: Error using activations with GoogLeNet in R2017b

The pretrained networks "alexnet" and "googlenet" belong to different MATLAB classes: "alexnet" is a SeriesNetwork, while "googlenet" is a DAGNetwork (where layers do not have to be arranged as one single chain). For a DAGNetwork, the "activations" method is not fully supported yet – this functionality will be available in a future MATLAB release.

But as you found out, the "activations" method does work on a "googlenet" to some extent. It returns a _h_-by-_w_-by-_c_-by-_n_ array, where _h_, _w_, and _c_ are the height, width, and number of channels for the output of the chosen layer, and _n_ is the number of observations.

The workaround is to reshape the matrix to be in the form of (number of observations) x (number of features) as is needed for "fitcecoc" for feature extraction. Here (number of features) = h*w*c. You should be able to reshape the training and testing matrices as follows:

>> nfeatures = size(trainingFeatures, 1) * size(trainingFeatures, 2) * size(trainingFeatures, 3);
>> nobservations = size(trainingFeatures, 4);
>> trainingFeatures = reshape(trainingFeatures, nfeatures, nobservations);
>> trainingFeatures = trainingFeatures';

For your reference, here are the documentation pages for SeriesNetwork and DAGNetwork:

Best Answer

Related Solutions

MATLAB: How to extract features from different layers in GoogLeNet

MATLAB: Error using activations with GoogLeNet in R2017b

Related Question