An Animal Classification based on Light Convolutional Network Neural Network
Bin Jiang, Wei Huang, Wenxuan Tu, Chao Yang · 2019
Designing an accurate and efficient model for animal recognition is a challenging task. It needs to consider many aspects including the accuracy of the model, the number of parameters, the complexity of calculation and so on. Therefore, we propose a novel convolutional network, called Bilateral Convolutional Network (BCNet), which aims at achieving a trade-off between recognition accuracy and model size, so that it can be better feasible to the mobile devices. It consists of two components, namely the feature extraction module and the classification module, respectively. The feature extraction module adopts a bilateral structure, which contains two sub-networks. A feature network for extracting image features and a location network for identifying object locations. The classification modeule adopts a two fully connected layer as a classifier. We tested on the Animals with Attributes animal (AWA) dataset, and the top-1 accuracy of the model was 85.6%. The model size of Bilateral Convolutional Network is 67.14 M, and computation complexity is only 1.12 GFLOPs. Compared with other methods, it achieves a good balance between recognition accuracy and model size.