Build-a-convolutional-neural-network-with-Tensorflow-and-Keras

This is a tutorial for building a CNN with tensorflow and keras. The model will be used for plate recognition. The model is then converted to used by OpenCV dnn in a C# application.

Prerequisite

• Packages: tensorflow, keras, scikit-learn, matplotlib
• Project structure

├───train_data
│   ├───0
│   ├───1
│   ├───2
│   ├───3
│   ├───4
│   ├───5
│   ├───6
│   ├───7
│   ├───8
│   ├───9
├───Convert_keras_to_tf.py
├───Prediction_tf_pb.py
├───train.py

• Link for the train data https://drive.google.com/open?id=1ls1J-F21J-Vzi9ZcgUL-_4QPhfoIdzBL

Implementation

• Train the model

python train.py --dataset train_data --model model.model --label-bin bin --plot plot

• Convert the model to tensorflow model

python Convert_keras_to_tf.py --keras_model model.model --tf_model tf_model.pb

• Test the tensorflow model

python Prediction_tf_pb.py

Result

• Accuracy: 99,9% after 20 epochs
• Processing time: 3ms/1 image

Acknowledgement

• The dataset is belong to Long Phung.