A Multi-scale Cattle Individual Identification Method Based on CMT Module and Attention Mechanism

Authors:

Abstract
With the development of precision animal husbandry technology, the automatic identification of individual cows has become essential. Traditional methods face limitations such as low efficiency, high cost, and vulnerability to environmental factors. This paper proposes a deep learning-based method for individual cow identification, integrating a CMT module and a multi-scale feature fusion branch atop the ResNet architecture. The CMT module captures long-range dependencies, enhancing the understanding of contextual information, while the multi-scale feature fusion branch improves feature expressiveness and robustness through hierarchical feature fusion. Additionally, an attention mechanism helps the model focus on critical features, boosting generalization, and a joint loss function combining circle loss and cross-entropy loss enhances identification accuracy. Experimental results show that this method achieves superior performance in cow identification, with a mean Average Precision (mAP) of 96.84%.

This publication has 21 references indexed in Scilit: