Plant Disease Classification with TensorFlow Lite on Android

At DevFest 2019 Yaounde , we created an end-to-end Android application with TFLite. We opted to develop an Android application that detects plant diseases.
Agenda :
1- Machine Learning overview
2- Building Deep Learning Model using TensorFlow and Keras
3- Deploying the Model to an Android App using TFLite

Transcript

1. Plant Disease Classification with TensorFlow Lite on Android By Yannick Serge Obam
2. - Software engineer & Co-Founder at isoMora - AI/ML Evangelist | Cloud Mentor | Teacher - Co-Organiser GDG Ebolowa Yannick Serge Obam @OBAMSerge
3. Agenda 1- Machine Learning overview 2- Building Deep Learning Model using TensorFlow and Keras 3- Deploying the Model to an Android App using TFLite
4. Machine Learning Overview
5. ML concepts Artificial Intellgence Machine Learning Deep Learning AI: The broad discipline of creating intelligent machines. ML: A system that can learn from experience. DL: A system that learn from experience on large data sets using deep neural networks.
6. Type of ML Supervised Learning • Data -> input • Label -> targets • Iteration of training, algorithm finds desired output • Regression and Classification problems Unsupervised Learning: • No labels • Not structured data • Clustering problems Reinforcement Learning: • Reward-based learning • learn from its experience
7. What’s a Neural Network? NN: a stack of layers. Layer: is made up of units . Unit: also called neurons, like the nerve cells in the biological brain. Dense Layer: stack of layers are fully connected Weight & Biases: internal variables of model which are updated during training Model: the representation of neural network
8. Step 1: Get Data Step 2: Preprocess the data Step 3: Choose an Algorithm Step 4: Train the model Step 5 : Run/Deploy the model
11. TensorFlow An open source Deep Learning library • >1,800 contributors worldwide • Built with C++ • Powers nearly every single ML project @ Google • Latest Release: 2.0 as of today TensorFlow 2.0 • Easier to learn and use • For beginners and experts
14. Deploy anywhere Servers Edge devices JavaScript
16. Build ML Model with TensorFlow U Use case : Plant Diseases Classification
17. Problem Statement The diagnosis of pests and diseases is essential for agricultural production. This lab invited participants to design algorithms and models to recognise species and diseases in the crop leaves
18. Images dataset 54 306- images of plant leaf for PlantVillage Dataset With 38 classes of combination Plant -> Diseases
19. Convolutional neural networks
20. @aaronhma http://bit.ly/plant-disease-model Demo
22. import time t = time.time() export_path = “/tmp/saved_models/{}”.format(int(t)) tf.keras.experimental.export_saved(model, export_path): export_path Export as saved Model
23. # convert the model to TFLite !mkdir ”tflite_models” TFLITE_MODEL = ”tflite_models/plant_disease_model.tflite” # Get the concrete function from the Keras model. run_model = tf.function(lambda x : reloaded(x)) # Save the concrete function. concrete_func = run_model.get_concrete_function( tf.TensorSpec(model.inputs[0].shape, model.inputs[0].dtype) Convert Model to TFLite
24. # Convert the model to standard TensorFlow Lite converter = tf.lite.TFLiteConverter.from_concrete_functions([concrete_func]) converted_tflite_model = converter.convert() open(TFLITE_MODEL, ”wp”).write(converted_tflite_model) Convert Model to TFLite
25. Deploy Model to an Android app
26. Download the Android skeleton app
27. 1- Open Android Studio. 2- Click Import project (Gradle, Eclipse ADT, etc.). 3- Choose the GreenDoctor/start/ folder from where you extracted the archive to. 4- Wait for the import processes to finish Import the app to Android Studio
28. Add TensorFlow Lite to the Android app Copy the TensorFlow Lite model plant_disease_model.tflite and plant_labels.txt into assets/ directory
29. Add TensorFlow Lite to the Android app
30. Add TensorFlow Lite to the Android app
31. @aaronhma http://bit.ly/plant-disease-android Dive into code
32. Finally you get this
33. Deploy and test app
36. Resources Blogs: Plant Disease Classification with TensorFlow Lite on Android Part 1 Plant Disease Classification with TensorFlow Lite on Android Part 2 Colab : http://bit.ly/plant-disease-model Source code Video: App demo Official Tensorflow site: https://www.tensorflow.org
37. Yannick Serge Obam @OBAMSerge
38. Yannick Serge Obam @OBAMSerge Thank you!