Comparison of ResNet50 and SqueezeNet1.1 for Plastic Bottles and Cans Classification

Waste problem is one of crucial global issue. Population growth, government regulations, people behavior and many other factors affect the waste management problem. This has a huge impact on the livability of the people in the neighborhood and maintaining a clean environment as well as managing waste is particularly challenging. Plastics and cans are the types of waste that are very difficult to recycle, so by using a reverse vending machine (RVM) these problems can be overcome. The critical point of RVM is the ability to recognize the type of waste. Deep learning has been successfully overperforms in many computer vision tasks, more specifically image recognition. Squezeenet and ResNet are two convolutional based deep learning algorithms. This study aims to compare the performance of two methods for plastic bottles and cans classification using SqueezeNet 1.1 and ResNet-50. From this research results, highest accuracy of SqueezeNet 1.1 is 92,5% with mean performance 1.518 sec and 97.5% for ResNet-50 wih 6.298 sec mean performance.