基于改进YOLO v8s的羊只行为识别方法

Sheep Behavior Recognition Method Based on Improved YOLO v8s

摘要:

羊只站立、行走、采食等日常行为与其健康状况密切相关，高效、准确的羊只行为识别有助于疾病检测，对实现羊只健康预警具有重要意义。针对目前羊只多行为识别检测大多基于传感器等接触式设备，羊只活动受限，行为具有局限性，且群体养殖环境下，羊只行为多样、场景复杂、存在遮挡等造成的行为识别精度低等问题，提出了一种基于改进YOLO v8s的羊只行为识别方法。首先，引入SPPCSPC空间金字塔结构增强了模型的特征提取能力，提升了模型的检测精度。其次，新增P2小目标检测层，增强了模型对小目标的识别和定位能力。最后，引入多尺度轻量化模块PConv和EMSConv，在保证模型识别效果的同时，降低了模型参数量和计算量，实现了模型轻量化。实验结果表明，改进YOLO v8s模型对羊只站立、行走、采食、饮水、趴卧行为平均识别精度分别为84.62%、92.58%、87.54%、98.13%和87.18%，整体平均识别精度为90.01%。与Faster R-CNN、YOLO v5s、YOLO v7、YOLO v8s模型相比，平均识别精度分别提高12.03、3.95、1.46、2.19个百分点。研究成果可为羊只健康管理和疾病预警提供技术支撑。

Abstract:

The daily behaviors of sheep, such as standing, walking, eating, drinking and sitting, are closely related to their health. Efficient and accurate recognition of sheep behaviors is crucial for disease and health detection. To address the current problem of the limited behavior of sheep caused by contact devices such as sensors and lower accuracy caused by diverse behaviors, complex scenarios, and occlusions in group farming, the method for sheep behavior recognition based on improved YOLO v8s was proposed. Firstly, the SPPCSPC was introduced to improve the feature extraction ability and the detection accuracy of the model. Secondly, the P2 detection was used to enhance ability of the model to identify and locate the small targets. Finally, multi-scale lightweight modules PConv and EMSConv were introduced and the number of parameters and calculation of the model were reduced and the lightweight was realized while ensuring the recognition of effects. The results showed that the average accuracy of the model proposed for standing, walking, eating, drinking, and sitting was 84.62%, 92.58%, 87.54%, 98.13% and 87.18%, respectively. And the overall average accuracy was 90.01%. Compared with Faster R-CNN, YOLO v5s, YOLO v7, and YOLO v8s model, the average accuracy was 12.03 percentage points, 3.95 percentage points, 1.46 percentage points, and 2.19 percentage points higher, respectively. The results can provide technical support for sheep health management and disease warning.