This repository contains the files and information related to the second project done by Lauren Smith, Ann Sofo and Reese Quillian in the UVA DS 4002 Project Course. In this project we explored the question: Can we differentiate between trash and recyclable materials through images?
The file in our src folder is a Jupyter notebook that contains all code used to clean our data, perform exploratory analysis, and build/evaluate our model. This notebook can be opened with any browser, but was originally created using Google Colaboratory. The following section outlines dependencies specific to Colab, but there may be others depending on the platform/browser you choose.
In order to successfully run the code chunks in the Jupyter notebook, a few packages need to be installed to supplement those that are pre-installed in the Colab environment. The lines to install these packages (!pip install python_splitter and !pip install tensorflow==2.7.0) can be found at the beginning of the notebook - just make sure these are run first. Then run the chunk that loads the libraries. Since Colab works in the cloud, the data is loaded from this github repository: once the repository is cloned, the required data files appear in the file system on the left hand side of the Colab environment. Once these steps have been completed, the notebook can be run and follows through all of our steps related to this project: data cleaning, exploratory data analysis, and building our (convolutional neural network) model.
The original dataset is a garbage classification dataset that was downloaded from Kaggle and was published in 2019. The six garbage classifications found in the data and their corresponding number of images are as follows: cardboard (393), glass (491), metal (400), paper (584), plastic (472) and trash (127). Due to the comparably smaller number of trash images found in this dataset, we decided to replace our current trash classification images with the trash and biological category of another dataset found on Kaggle to even out our classes, which left us with 1,682 images of trash. In total we have 3,102 observations of garbage images, which we separated into two classes before exploring our research question: recyclables (2,390), trash (1,682). The recyclable class was formed from grouping together the other five categories aside from trash.
| Folder | Description |
|---|---|
| recycle | 2390 JPG images of recyclable waste, including images of paper, plastic, metal, cardboard, and glass |
| trash | 1682 JPG images of trash, including biological waste |
| Figure | Takeaway |
|---|---|
| Validation Accuracy Graph | After the model was trained for three epochs, it was determined that the accurary for the testing dataset was about 93%. Overfitting was avoided as the testing accuracy was higher than the training accuracy. |
| Test & Training Splits | Used to verify that the classes were appropriately balanced after being split into testing and training. |
| Sample Images | These depict random samples from the recycling and trash image folders. These were used to verify that the images were an appropriate quality after being resized as part of our data cleaning process. |
[1] D. Rausch, “EDA for Image Classification,” Geek Culture, Apr. 15, 2021. https://medium.com/geekculture/eda-for-image-classification-dcada9f2567a (accessed Mar. 16, 2023).
[2] Chang, “Garbage Classification.” https://www.kaggle.com/datasets/asdasdasasdas/garbage-classification (accessed Mar. 16, 2023).
[3] Mostafa Mohamed, “Garbage Classification (12 classes).” https://www.kaggle.com/datasets/mostafaabla/garbage-classification (accessed Mar. 16, 2023).
[4] N. Lang, “Using Convolutional Neural Network for Image Classification,” Medium, Apr. 28, 2022. https://towardsdatascience.com/using-convolutional-neural-network-for-image-classification-5997bfd0ede4 (accessed Mar. 16, 2023).
[5] B. Adhikari, “Python Splitter,” GitHub, Mar. 14, 2023. https://github.com/bharatadk/python_splitter (accessed Mar. 16, 2023).
[6] A. H, “Image classification using CNN for Beginners,” Kaggle, 27-Aug-2021. [Online]. Available: https://www.kaggle.com/code/anandhuh/image-classification-using-cnn-for-beginners/notebook. (Accessed: 21-Mar-2023).
[7] J. McDermott, “Convolutional Neural Networks - image classification W. Keras,” Learn Data Science - Tutorials, Books, Courses, and More. [Online]. Available: https://www.learndatasci.com/tutorials/convolutional-neural-networks-image-classification/. (Accessed: 21-Mar-2023).
[8] K. Team, “Keras documentation: The Sequential class,” keras.io. https://keras.io/api/models/sequential/. (Accessed: 21-Mar-2023).
[9] N. Lang, “Using Convolutional Neural Network for Image Classification,” Medium, Apr. 28, 2022. https://towardsdatascience.com/using-convolutional-neural-network-for-image-classification-5997bfd0ede4. (Accessed: 21-Mar-2023).
[10] [1]“Overfitting in CNNs | Learn different ways to Treat Overfitting in CNNs,” Analytics Vidhya, Sep. 07, 2020. https://www.analyticsvidhya.com/blog/2020/09/overfitting-in-cnn-show-to-treat-overfitting-in-convolutional-neural-networks/. (Accessed: 32-Mar-2023).