Desarrollo de un detector del ave Huiracchuro en cultivos mediante el reconocimiento de imágenes basado en deep learning

Abstract

This thesis focuses on the development of a real-time detection system for the Huiracchuro bird in crops using deep learning techniques. The pre-trained MobileNetv2 neural network was utilized, which was modified to specifically detect this bird species. The modification involved removing the final classification layer and adding a new layer to identify the "Huiracchuro" class when it is detected through the camera of a mobile device. The implementation was carried out in a web application, enabling users to access the network and use their device cameras for real-time detection. NGROK was employed to establish communication between the web application and the local server hosting the trained neural network. The system was evaluated through field tests conducted in the metropolitan district of Quito, yielding promising results. The neural network achieved an accuracy of over 70% in detecting the Huiracchuro bird, provided that it occupied a minimum area of 40x40 pixels in a 224x224 pixel image. This work represents a significant advancement in the real-time detection and recognition of birds in crops using deep learning. The real-time focus and availability on mobile devices offer a practical and accessible solution for farmers. Moreover, the utilization of the MobileNetv2 neural network, Python-based network modification, and training on the Google Colaboratory platform demonstrate the effectiveness and feasibility of the methodology employed. In conclusion, this real-time detection system for the Huiracchuro bird in crops provides a valuable tool for plant conservation, crop protection, and the economic income of farmers. With an accuracy exceeding 70%, this system showcases its potential for field applications.