[Paper Review] Comparative Analysis of Multiple Deep CNN Models for Waste Classification

Dipesh Gyawali, Alok Regmi+3 more|arXiv (Cornell University)|Apr 5, 2020

Advanced Neural Network Applications8 references31 citations

TL;DR

The paper compares multiple pre-trained CNNs (notably ResNet18) for waste image classification using transfer learning, achieving best validation accuracy of 87.8% with ResNet18. It analyzes feature extraction, fine-tuning, and model performance on a 4-class waste dataset.

ABSTRACT

Waste is a wealth in a wrong place. Our research focuses on analyzing possibilities for automatic waste sorting and collecting in such a way that helps it for further recycling process. Various approaches are being practiced managing waste but not efficient and require human intervention. The automatic waste segregation would fit in to fill the gap. The project tested well known Deep Learning Network architectures for waste classification with dataset combined from own endeavors and Trash Net. The convolutional neural network is used for image classification. The hardware built in the form of dustbin is used to segregate those wastes into different compartments. Without the human exercise in segregating those waste products, the study would save the precious time and would introduce the automation in the area of waste management. Municipal solid waste is a huge, renewable source of energy. The situation is win-win for both government, society and industrialists. Because of fine-tuning of the ResNet18 Network, the best validation accuracy was found to be 87.8%.

Motivation & Objective

• Motivate automatic waste sorting to aid recycling and reduce human labor.
• Evaluate transfer learning with pre-trained CNNs for waste image classification.
• Identify the best performing model and analyze its learning behavior and confusion patterns.

Proposed method

• Use transfer learning with pre-trained CNNs (ResNet and VGG) on a 4-class waste dataset (Plastic, Paper, Metal, Glass).
• Preprocess images: resize to 512x512, normalize channels, and augment data to improve generalization.
• Unfreeze last convolutional layers for feature extraction and then fine-tune with added classifier layers.
• Split data 80/20 for training/validation; optimize with Adam (lr=0.001) and negative log-likelihood loss.
• Train on an NVIDIA GTX 1050 with PyTorch to compare model performance and learning curves.
• Select ResNet18 as best performing model after comparing training/validation loss and accuracy.

Experimental results

Research questions

• RQ1 Which pre-trained CNN architectures (ResNet, VGG, etc.) perform best on the waste classification task?
• RQ2 Does transfer learning with fine-tuning improve waste classification accuracy on the provided dataset?
• RQ3 How do training and validation losses and accuracies evolve for each model, indicating overfitting or underfitting?
• RQ4 What are the misclassification patterns across waste categories, and what do they reveal about dataset challenges?

Key findings

• Best validation accuracy achieved with ResNet18 at around 87.8%.
• ResNet50 and VGG16 showed overfitting tendencies with diverging training/validation losses.
• ResNet18 demonstrated decreasing training and validation losses with reasonable validation accuracy and no significant overfitting.
• Confusion matrix reveals higher misclassifications between Paper and Plastic and that Glass is often mistaken for Plastic, while Metal predictions are relatively strong.
• Feature maps illustrate progressive feature extraction from edges to more complex shapes across CNN layers.
• Data augmentation and normalization were used to mitigate overfitting and improve generalization.