Waste Classification Android App

CAPSTONE PROJECT by C22PC385

Get the apk here.

How to Use The App

1. Open the app

2. Click the camera button on the dashboard page

3. Take picture of waste object

4. Click the predict button

5. See the result

Machine Learning Part

Download the dataset we used here.
See the Jupyter Notebook here.
Content of Jupyter Notebook:

1. Import libraries
2. Import dataset
   The dataset contains of 11 classes of images, which are battery, cardboard, clothes, e-waste, food, glass, light bulbs, metal, paper, plastic, and shoes.
3. Image preprocessing using ImageDataGenerator
4. Build the model structure with InceptionV3 pre-trained model and addition layers (dense layer with an output size of 11 and activation of softmax as the last layer)
5. Compile and train the model
6. Plot validation accuracy and loss
7. Model testing
8. Save model into .h5 format and TFLite (additional)
9. Load the .h5 format saved model and retrain it

Cloud Computing Part

1. Activate Cloud Run API and Cloud Build API
2. Container image to package resources (model.h5, Flask RestFul API) using Dockerfile
3. Build with Cloud Build and deploy it to Cloud Run
4. API endpoints: https://getpredict-ehmfuclc5q-et.a.run.app
5. The mapping prediction results into 3 categories:
   Organik (Organic): food, cardboard, paper
   Anorganik (Inorganic): clothes, glass, light bulbs, metal, plastic, shoes
   B3 (Toxic and Hazardous Material): battery, e-waste
   

Mobile Development (Android) Part

1. Capture image using CameraX
2. Call the endpoint API with the image as a key
3. Display the result