A Prototype Study on YOLOv10-Based Bird Gesture Recognition

Abstract

Birds are one of the most abundant types of creatures on Earth. However, it is also known that there are a large number of taxonomically diverse bird species in nature. The bird network has standard behavioural patterns such as flying, perching, feeding and walking. In this study, 2372 bird images are used for five standard bird gestures detection which are flying, perching, swimming, eating, and walking with the Yolov10 algorithm from Caltech-UCSD Birds-200-2011 dataset. Firstly, the dataset is prepared for detection by classifying these gestures. Secondly, the bird gesture images are trained with Yolov10, thirdly the trained model is tested with bird motion short videos and finally, the evaluation results are shown with evaluation metrics. In this prototype study, it was observed that the obtained model had results with accuracy higher than 70%. The study can be used to make sense of bird communication for future studies.