Year: 2024
Type: conference-paper
Abstract: Birds, as important ecological indicators, their diversity and abundance changes directly reflect the health of the environment. This paper conducts research and analysis on the fine-grained recognition problem of birds. In the context of bird diversity, the recognition is not limited to the positioning of the head and torso. We propose a classification method based on Faster RCNN for joint semantic extraction and detection. An end-to-end network is constructed, utilizing a top-down geometric restriction method with K-NN to generate seven regional candidate locations. Feature extraction is performed on the candidate regions according to constraint conditions. This method significantly reduces the number of candidate regions at an order of magnitude. Compared to other bird recognition research, the recognition accuracy in our experiments has increased to 88.37%. The results indicate that the proposed method is suitable for fine-grained bird recognition, achieving discrimination of bird diversity.
Author: Chuanjin Nie
Institution: Wuhan Donghu University
Language: English
Subfield: Computer Vision and Pattern Recognition
Field: Computer Science
Domain: Physical Sciences
Open Access status: bronze
OpenAlex ID: W4400913881