Model

Matlab Classify Image Using GoogLeNet see here

Test

Use Alexnet accuracy 97.52%, Googlenet accurarcy 99.44, resnet50 accuracy 89.64% classify 1000 different categories.

Type

We only need six different types of transportation image to be validated

Modify and Training process

1.Delete matlab provided model with last 3 layers.

prevent overfitting

2.connect fully-connected layer, softmax layer, classification layer.

reduce model size and prevent overfitting

3.set imageAugmenter = imageDataAugmenter(...) mirroring, translation for imageDatastore to create dataset for augumented Train, validation.

4.use Inputdata as 80% for train datastore, 20% for validation datastore.

5.setting trainingOptions : Initial learning rate, max epochs, mini batchsize.

6.start use net_Train = trainNetwork(augumented training datastore, layers, trainingoptions) to get train model(or net).

7.use 20% validation datastore and the net just trained, with classify(net, validationdatastore) to get prediciton scores.

8.use mean(Ypred == imdsValidation.Labels) to get accuracy.

9.show the result below.

Target: This code is for objects recognition , you can use kaggle dataset. Enviroment: Matlab Training model: googlenet_matlab.mat

detail: validation_cnn.m is validation program. Do: Just change imagedatastore 's parameter to your test image folder(input image) path.

other things you can do: You can create another objects for training like cat, dog ... ex. ../validation/cat/.. or ../validation/dog/..