Efficient Pig Counting Method Based on Improved YOLOv10n

Counting pigs within a designated area is a fundamental yet challenging task in pig farm management. Manual counting is not only time-consuming and labor-intensive but also prone to errors, resulting in low efficiency. Automatic pig counting using deep learning can work even better. Considering the computing power of low-cost deployed devices in practical applications, it is necessary to achieve better results with limited computing power. To address this issue, we propose a pig detection and counting method leveraging deep learning, specifically tailored for resource-constrained deployment environments. Our approach is a lightweight solution, built upon the foundation of YOLOv10n, with key modifications. We introduce the Faster Block to construct the C2f-FN module within the backbone network, effectively reducing computational complexity and parameter count. Additionally, an attention layer is appended to the end of the backbone to maintain high accuracy, incorporating the lightweight MLCA attention mechanism for enhanced channel and spatial information extraction. In the Neck section, we discard the conventional upsampling structure and adopt the lightweight yet efficient DySample upsampler, achieving a balance between model precision and lightweight design. The proposed method has a good lightweight level and detection accuracy. The verification results on the self-built data set show that the proposed method exceeds the accuracy of the original YOLOv10n version, and the calculation amount and parameter number are reduced by 23.1% and 21.1% compared with the original YOLOv10n version.