[Paper Review] Fine-Tuning Models Comparisons on Garbage Classification for Recyclability

Umut Özkaya, Levent Seyfi|arXiv (Cornell University)|Aug 7, 2019

Advanced Neural Network Applications7 references47 citations

TL;DR

The paper evaluates fine-tuned CNN architectures (AlexNet, VGG16, GoogLeNet, ResNet) with Softmax and SVM classifiers for garbage type classification using a 2527-image dataset, achieving up to 97.86% accuracy with GoogleNet+SVM.

ABSTRACT

In this study, it is aimed to develop a deep learning application which detects types of garbage into trash in order to provide recyclability with vision system. Training and testing will be performed with image data consisting of several classes on different garbage types. The data set used during training and testing will be generated from original frames taken from garbage images. The data set used for deep learning structures has a total of 2527 images with 6 different classes. Half of these images in the data set were used for training process and remaining part were used for testing procedure. Also, transfer learning was used to obtain shorter training and test procedures with and higher accuracy. As fine-tuned models, Alexnet, VGG16, Googlenet and Resnet structures were carried. In order to test performance of classifiers, two different classifiers are used as Softmax and Support Vector Machines. 6 different type of trash images were correctly classified the highest accuracy with GoogleNet+SVM as 97.86%.

Motivation & Objective

• Develop a deep learning application to classify garbage types from images for recyclability.
• Assess performance of multiple fine-tuned CNN architectures on a 6-class garbage dataset.
• Evaluate two classifiers (Softmax, Support Vector Machines) with transfer learning.
• Identify the best combination of model and classifier for high accuracy.

Proposed method

• Use transfer learning with AlexNet, VGG16, GoogLeNet, and ResNet on a 6-class garbage dataset (2527 images).
• Split dataset into 50% training and 50% testing.
• Compare Softmax and SVM as the final classifiers.
• Report accuracy of each model-classifier pair to identify the best performing configuration.

Experimental results

Research questions

• RQ1 Which fine-tuned CNN architecture yields the highest classification accuracy for garbage recyclability?
• RQ2 Does using an SVM classifier improve performance over Softmax for these fine-tuned models?
• RQ3 How does transfer learning affect training efficiency and accuracy on a small dataset of garbage images?
• RQ4 What is the best model-classifier combination for 6 garbage categories?
• RQ5 What is the overall feasibility of a vision-based garbage type detector for recyclability?

Key findings

• GoogleNet paired with SVM achieved the highest accuracy of 97.86%.
• The study demonstrates that transfer learning can shorten training and testing procedures while maintaining high accuracy on a small dataset.
• Among the tested architectures, AlexNet, VGG16, GoogLeNet, and ResNet were evaluated with both Softmax and SVM classifiers.
• Two classifiers (Softmax and SVM) were used to assess their impact on final performance across models.
• The dataset consisted of 2527 images across 6 garbage classes, with a 50/50 train/test split.