Waste Classifications Using Convolutional Neural Network
Hajar Abdulrahman, Nabil M. Hewahi · 2021 International Conference on Data Analytics for Business and Industry (ICDABI) · 2021
This research presents the use of automated machine learning models to manage and separate waste efficiently. The focus of this study is to use Convolutional Neural Network (CNN) with 3 different architecture designs to compare which gave the best model. Moreover, this research further investigates the model’s performance when using RGB images and greyscale images. The dataset used was obtained from Kaggle website that contains 156,362 images belonging to two classes: biodegradables and non-biodegradables. The best model achieved adequate results were of architecture 3 using the greyscale images with accuracy 81.0% and reported loss of 44%. Similar results obtained with RGB image using the same architecture with accuracy of 82.7% and loss 49%. The greyscale images showed better performance for all model structures, where it reported loss value less than the RGB images, and similar accuracy. Moreover, the impact of FC layers in CNN for image classification was positive, it enhanced the models performance. The moderate accuracy and extreme high loss value were due to the overfitting of the low-quality images. Future study can add other hyper parameter to increase accuracy such as more dropout layers on a good quality data, and to further understand the impact of FC layers in architecture.