Monitoring based on computer vision of endangered birds in the Pantanos de Villa reserve

Abstract

Governments, along with numerous conservation organizations, work to protect birds primarily in nature reserves, specially in the Pantanos de Villa reserve in Peru. Therefore, it is necessary to carry out permanent censuses to monitor these populations, in an agile way and at a low cost, without requiring the intervention of a human expert. The present work proposes a monitoring system consisting of a detection model using Mask R-CNN with a public Open Images Dataset V6 dataset and a Deep Convolutional Neuronal Network classifier with a dataset of four bird species from the Pantanos de Villa reserve, Peru. The system will be used to monitor images captured by the workers themselves, where the detection and classification are carried out; After several daily evaluations, the existing population of the birds will be
determined, and when there is a decrease, the reserve will take action. The proposed system managed to detect and classify bird species for monitoring, obtaining an accuracy of 82,49%, improving the values previously obtained using only the classifier.