Research on Small Sample Bird Recognition Based on B-CNN
Download as PDF
DOI: 10.23977/iset.2019.045
Author(s)
Qiang Gao, Yutong Chen, Jiaxing Lei and Xianyu Lu
Corresponding Author
Qiang Gao
ABSTRACT
Bird recognition is a typical fine-grained image classification task. Bilinear-CNN (B-CNN) is a widely used fine-grained image recognition framework. Based on the Object Detection API in the TensorFlow framework, this paper combines the B-CNN algorithm, uses data set fusion to achieve data amplification, and then performs secondary training to solve the problem of small sample training difficulties. Finally, the high-risk airport is realized by B-CNN. Automatic recognition of bird images. The experimental results show that the image neural network detection model based on B-CNN can fully utilize the flexibility of TensorFlow and B-CNN, and effectively improve the stability, speed effectiveness and accuracy of image recognition. The average detection accuracy reaches 85.8. %.
KEYWORDS
Bird Recognition, B-CNN, fine-grained image