Propuesta de identificación de residuos sólidos urbanos para la clasificación en la fuente mediante la técnica de visión artificial

Abstract

This research proposes the identification and classification of municipal solid waste at the source using artificial vision, aligning to the regulatory classification methodology of resolution 2184 of 2019. A convolutional artificial neural network (ANN) was developed using Teachable Machine, structured in two layers, with an input of 224×224 pixels and 3 RGB channels. The model processed 2600 training images and achieved an accuracy of 80% and a sensitivity of 71%. To optimize classification, labels were grouped according to normative and visual criteria, reducing the categories to 11 labels or classes. The graphical user interface (GUI) facilitated user interaction by allowing real-time identification via camera or analysis of stored images. This solution represents a breakthrough in the management of urban solid waste in Colombia, promoting its correct classification and improving its reincorporation into the circular economy, thus contributing to environmental sustainability.