Programa del congreso

— First, rapid urban growth and rising consumption in Peru have intensified the logistical and public-health challenges associated with municipal solid waste (MSW) management. Moreover, the segregation of recyclable waste is still largely performed manually, exposing operational workers to biological hazards and limiting the efficiency of treatment plants. In this context, computer vision and convolutional neural networks (CNNs) provide promising alternatives for automating waste classification; however, performance often degrades in local deployments when models are trained exclusively on foreign datasets that do not capture the visual characteristics of products in the Peruvian market (e.g., local brands). Accordingly, this work proposes an automatic waste classification system tailored to the Peruvian context using transfer learning techniques. To this end, a hybrid dataset was built by integrating the standard TrashNet database with a new corpus of images of Peruvian household waste. Finally, 20 experimental configurations were implemented and evaluated, comparing deep architectures (ResNet50, Xception) against lightweight models (MobileNetV2) and baseline variants under different optimizers and data augmentation strategies. Preliminary results suggest that domain adaptation through local data significantly increases classifier accuracy, supporting the technical feasibility of intelligent recycling systems for Peru.