ASAH: Machine Learning

⚠️ About Our App

ASAH (Aplikasi Sortir Sampah) is an application to encourage people to manage their waste properly by giving them rewards if they successfully manage their waste properly, where they need to sort their waste first and then they can dispose of or give their waste to recycling agencies around them

📚 Related Project Repositories

Here are some of the related repositories which are part of the same project:

Repository	Link
📱 Mobile Development	MD Repository
☁️ Cloud Computing	CC Repository

🤖 Machine Learning: Image Classification for Waste Sorting

This project utilizes machine learning to facilitate the waste sorting process. We have developed an image classification model that accurately identifies the type of waste in a given image. The model, based on MobileNetV2, is trained to categorize waste into 10 distinct classes.

The classes are:

1. Shoes
2. Metal
3. Plastic
4. Glass
5. Clothes
6. Paper
7. Trash
8. Battery
9. Biological
10. Cardboard

🌐 Overview

Here are some snapshots of the project:

📈 Model

In this project, we explored two different Convolutional Neural Network architectures - VGG16 and MobileNetV2.

• VGG16: We first implemented the VGG16 model. However, we only achieved a validation accuracy of 87.9%.

• MobileNetV2: Switching to MobileNetV2, we were able to improve our validation accuracy significantly, achieving a score of 94%.

The superior performance of MobileNetV2 can be attributed to its efficient architecture that uses depth-wise separable convolutions to reduce the model size and complexity, which is particularly important for mobile applications.

📚 Libraries Used

This project utilizes several Python libraries for data handling, machine learning, and visualization:

Library	Purpose
os	Provides functions for interacting with the operating system.
shutil	Used for high-level file operations.
zipfile	Allows the reading and writing of ZIP-format archives.
pathlib	For manipulating filesystem paths.
random	Generates random numbers, selects random elements from lists.
cv2	OpenCV for image and video processing.
numpy	Enables numerical computing with powerful numerical arrays objects, and routines to manipulate them.
tensorflow	An open-source platform for machine learning.
matplotlib.pyplot	Used for creating static, animated, and interactive visualizations in Python.
PIL (Python Imaging Library)	Adds image processing capabilities to your Python interpreter.
os	Provides a portable way of using operating system dependent functionality.
keras.utils	Contains various utility functions for Keras.

📝 Documentation

Data Acquisition

The data for this project was sourced from Kaggle. Initially, the dataset comprised of 12 classes of waste, represented through a total of 15,515 images. However, we decided to combine some categories to make a total of 10 distinct waste classes for ease of classification and practicality in waste sorting.

Model Training

We used TensorFlow to train our image classification model. The MobileNetV2 architecture, pre-trained on the ImageNet dataset, serves as the backbone of our model. This enables the model to extract useful features from the waste images and classify them accurately.

Using the Model

The trained model can be used by ASAH app users to classify their waste. Users simply need to take a picture of the waste item, and the app will tell them which category it belongs to. This assists users in sorting their waste correctly, leading to more efficient recycling and waste management.

⬇️ Download the Model

The trained model can be downloaded for your own use and further development. Click here to access and download the model.

💭 Prediction

In this section, we showcase the use of the trained model for predicting the type of waste in an image. First, the model is loaded from a saved file. Then, an image of waste is uploaded and preprocessed to the appropriate size for input into the model. The model predicts the class probabilities, and the most likely class is determined by finding the class with the highest probability. The image of the waste along with the model's prediction accuracy is then displayed.

Here's the Python code:

# Load the model from the saved file
load_model = tf.keras.models.load_model(path)

# Get the filename of the uploaded image
fn = list(uploaded.keys())[0]
path = fn

# Load and preprocess the image
img = image.load_img(path, target_size=IMG_SIZE + (3,))
x = image.img_to_array(img)
x = np.expand_dims(x, axis=0)

# Make a prediction using the model
images = np.vstack([x])
classes = load_model.predict(images, batch_size=10)
outclass = np.argmax(classes)

# Display the image and prediction accuracy
plt.imshow(img)
plt.axis('off')

accuracy_percentage = classes[0][outclass] * 100 
plt.title(f"{accuracy_percentage:.2f}%")
plt.show()

🏆 Credits

This project utilizes a dataset from Kaggle: Garbage Classification (12 classes) uploaded by MOSTAFA MOHAMED.