This project aims to classify banana images according to their maturity using a deep learning model. It involves data preprocessing, model building using TensorFlow and Keras, class balancing, hyperparameter tuning and saving of the trained model.
For this banana ripening process classification project, I used a dataset provided by Roboflow. The dataset includes images of bananas at various stages of ripeness, labeled with categories such as freshripe, freshunripe, overripe, ripe, rotten, and unripe. You can find the original dataset https://universe.roboflow.com/fruit-ripening/banana-ripening-process.
P.S.: You can use any dataset of 2 classes and up, you only have to refactor the code up to a certain point and then the process is the same in this case.
To run this project, make sure you have the following dependencies installed:
- Python 3
- TensorFlow
- Keras
- pandas
- numpy
- PIL
- matplotlib
P.S.: It can also be executed in Google Colab with BananaRipnessProcess.ipynb which is available in the repository.
git clone https://github.com/MaitreStick/bananaRipeningProcessFollow these steps to use the project:
- Set up the required libraries and dependencies.
- Run the data preprocessing script to clean and organize the dataset.
- Build the deep learning model using TensorFlow and Keras.
- Train the model using the prepared dataset.
- Evaluate the model performance and make necessary adjustments.
- Save the trained model for future use.
Since I used only 5 classes of the 6 that the dataset offers and I balanced the classes so that they were all with the same number of elements, the percentage of accuracy was 62.2%, however in roboflow there are some examples where more than 90% of accuracy is obtained, this is only another way to do it. The trained model is available in the repository for reference.
P.S.: I should also say that the training took us 8 hours in total, as the best hyperparameter to do the training with was first searched for.