Identification of Araucaria angustifolia trees in an urban forest fragment using UAV images and YOLOv7 structure

Abstract This study addresses the identification of individuals of the Araucaria angustifolia species in urban forest fragments, specifically in the Mixed Ombrophilous Forest (FOM) in Curitiba, Paraná, Brazil. The aim of the study is to use UAV images and the computer vision technique of the YOLOv7 model to detect individuals of A. angustifolia. The FOM is essential for local biodiversity conservation and human well-being but faces challenges due to urban sprawl and the conversion of land use to agriculture. The species is critically endangered, requiring actions and strategies for its conservation. The study highlights the role of Unmanned Aerial Vehicles (UAVs) and deep learning techniques, such as Convolutional Neural Networks (CNNs), in identifying tree species in urban ecosystems. YOLOv7, an architecture based on CNNs, was chosen because of its detection capacity. YOLOv7 is especially effective at detecting a wide variety of objects, including people, vehicles, animals, household objects, road signs and much more, making it an ideal choice for identifying tree species in urban environments. The data was obtained by a DJI Mavic 3 UAV. Utilizing a UAV, the study area of the urban forest was flown over, generating an orthomosaic that was subsequently divided into 14 parts for training, validation, and testing. The YOLOv7 model was trained with the images to detect A. angustifolia trees present in the area. The results show that model achieved a precision of 79.3%, recall of 86.8%, and Mean Average Precision of 87% during training. Comparative analysis with forest inventory data reveals promising performance in detecting A. angustifolia trees. The average confidence of the model's classification was 76.18 ± 12.88%, with 80.81% being the most frequent classification for the median result. The present study uses the effective integration of UAV technology, YOLOv7 model with deep learning technique to detect and assess tree species in urban ecosystems. This approach provides an important tool for conservation strategies aimed at assessing and managing the tree biodiversity in urban forest remnants.