Desarrollo de un algoritmo de visión artificial para la clasificación de las especies de patos dendrocygna bicolor y dendrocygna autumnalis en zonas aeroportuarias de Guayaquil

Abstract

This project designs and implements a computer-vision system to detect two whistling-duck species, Dendrocygna autumnalis and Dendrocygna bicolor, aimed at supporting monitoring in sensitive environments. A dataset was built from public images gathered on the web, labeled by species with LabelImg, and organized into training, validation, and test partitions. On this basis, a lightweight YOLOv7-tiny model was trained in Google Colab via transfer learning and exported to ONNX for execution on a Raspberry Pi connected to a UVC-compatible Full HD webcam. The solution integrates image acquisition, inference, and display visualization through bounding boxes, achieving low latency and stable operation. The system exhibits robust behavior across diverse scenes and a clear separation between species. As a practical contribution, the work provides a reproducible end-to-end pipeline, device-ready inference scripts, and calibration guidelines that facilitate field use and extension to new scenarios.