Carbon/Nitrogen Ratio Estimation for Urban Organic Waste Using Convolutional Neural Networks

Abstract

In this paper, the Carbon/Nitrogen ratio was estimated by classifying the urban organic waste (UOW) based on qualitative (color and maturity) and quantitative (weight) characteristics via convolutional neural networks (CNN) and image processing. The reuse of UOW is a suitable process in waste management, preventing its disposition in landfills and reducing the effects on the environment and human health. Ambient conditions affect the UOW characteristics over time. Knowing these changes is essential to reuse them appropriately, mainly both carbon and nitrogen content. A categorization associated with the decomposition stage of the UOW was proposed, which becomes the corresponding UOW classes. Three convolutional neural network models were trained with UOW images. Two pre-trained CNN (MobileNet and VGG16) were trained by transfer learning technique, and one proposed model (UOWNet) was trained from scratch. The UOWNet model presented a good agreement for the classification task. The results show that this preprocess is a practical tool for assessing the Carbon/Nitrogen ratio of UOW from its qualitative and quantitative features through image analysis. It is a preliminary framework aimed to support household organic waste recycling and community sustainability.