基于深度学习的家畜虚拟电子围栏设计

针对牧场中家畜越界及传统电子围栏难以防止家畜越界的问题，基于改进后的YOLOv5，设计了虚拟电子围栏。首先采用YOLOv5s模型作为基础，并进行了迁移学习和添加ECA注意力模块。然后利用PyTorch框架进行训练，并对模型进行了评估，相比原版YOLOv5s，改进YOLOv5s对黄牛的检测精确率、召回率、mAP分别提升0.2、1.3、0.7个百分点，单帧推理总耗时下降0.5ms。最后将改进后的模型转换为RKNN格式，并部署在带有NPU的RK3588开发板上，加快模型推理速度。结果表明，通过深度学习技术与ROI划定技术的应用，成功设计了家畜虚拟电子围栏，优化智慧牧场的管理体系，提高管理效率，降低管理成本，并具备一定的实用价值。

Design of Livestock Virtual Electronic Fence Based on Deep Learning

In response to the issue of livestock crossing boundaries in pastures and the limitations of traditional electronic fences in preventing such occurrences, a virtual electronic fence was designed based on the improved version of YOLOv5. Initially, the YOLOv5s model was used as the foundation, and transfer learning and the addition of an ECA attention module were performed. Subsequently, the model was trained using the PyTorch framework and evaluated. Compared to the original YOLOv5s, the improved YOLOv5s achieved a 0.2%, 1.3%, and 0.7% increase in precision, recall, and mAP for cattle detection, respectively, while reducing the single-frame inference total time by 0.5 ms. Finally, the improved model was converted to the RKNN format and deployed on an NPU-equipped RK3588 development board to accelerate the model''s inference speed. The result demonstrate the successful design of a livestock virtual electronic fence through the application of deep learning techniques and ROI delineation. This system optimized the management system of smart farms, enhanced management efficiency, reduced costs, and proved to be of practical value.