Fine-grained classification of fly species in the natural environment based on deep convolutional neural network. (August 2021)

• Record Type: Journal Article
• Authors: Chen, Yantong; Zhang, Xianzhong; Chen, Zekun; Song, Mingzhu; Wang, Junsheng
• Is Part Of: Computers in biology and medicine. Volume 135(2021)
• Publication Date: 2021-08
• Subjects: Deep convolutional neural network -- Fly species classification -- Feature fusion -- Fine-grained classification -- Deep learning; Medicine -- Data processing -- Periodicals; Biology -- Data processing -- Periodicals; 610.285
• DOI: 10.1016/j.compbiomed.2021.104655
• Languages: English
• ISSNs: 0010-4825

Abstract

Abstract: Effective classification of flies is beneficial to prevent the spread of disease and protect agricultural production. It is important to prevent the invasion of fly species. Aiming at the problem of similar morphology and difficulty in the classification of fly species in the natural environment, this paper proposes a fine-grained classification method for fly species in the complex natural environment based on deep convolutional neural network. Firstly, the specific position of the fly in the image is located by the gradient-weighted class activation graph method, and the object region of the fly is obtained. Then, the local region containing the most abundant information in the image is extracted. When extracting features from the local region, the attention module and cross-layer bilinear pooling are combined. The feature information of different convolutional layers is integrated. Finally, the global and local feature information is integrated for classification. We experimentally compared the proposed method with other state-of-the-art methods on the established dataset. Experimental results show that the accuracy of the proposed method on the three datasets is 84.34%, 89.53% and 93.26%, respectively. Compared with other state-of-the-art methods, this method has a good classification effect on fly species. Highlights: Fine-grained classification of fly species in the natural environment. Classification based on object region localization and feature fusion network. The network combines global features and local features for classification. The model in this paper has a good classification effect on fly species.