DCAM-Net: Sheep Gender Identification Network Based on Dilated Convolutional Attention Module
Xinyu Zhang, Tian Zhenzhen, Wei Yan, Lyu Lei, Wang Jihua · 2023
Automatic identification of the genders of sheep can be valuable to the sheep industry. Sheep producers need to identify the genders of sheep to forecast the population variation of their flock and combine it with some research works on sheep faces. Yet, in many cases, farmers find it difficult to distinguish the gender of sheep when do not have a great deal of experience. But, recent advancements in deep learning in computer vision will help to classify sheep's gender quickly and accurately. In this paper, we propose a novel sheep gender identification network based on dilated convolutional attention module (DCAM-Net). In the framework, multi-stage and multi-scale feature fusion effectively captures multi-scale features to resist noise interference and scale changes. Additionally, we design a channel and spatial attention module called the dilated convolutional attention module (DCAM) to promote the network to better focus on foreground information and suppress background information. There is no publicly available dataset with enough data for gender recognition in deep architectures. Therefore, to address the problem, we build a dataset of 8768 sheep images of 428 rams and 420 ewes acquired on a farm and annotated by an expert. We test the performance of DCAM-Net on the sheep face dataset, and the experimental results prove the superiority of the proposed method and the attention module.