基于YOLOv5n-ChtaFSD模型的飞机草精准检测

Precision detection of Siam weed Chromolaena odorata based on a YOLOv5n-ChtaFSD model

中文关键词:飞机草  目标检测  YOLOv5模型  特征融合  改进

keywords:Chromolaena odorata  object detection  YOLOv5 model  feature fusion  improvement

中文摘要:为解决飞机草Chromolaena odorata识别效率低、成本高的问题，基于YOLOv5n模型构建了一种改进的目标检测模型YOLOv5n-ChtaFSD，该模型在颈部网络引入重参数化泛化特征金字塔网络结构，该结构通过消除冗余的上采样操作来提升精度，降低推理耗时；该模型在颈部网络嵌入简单注意力机制，该机制根据特征重要性实现自适应筛选，从而增强模型对小目标的感知能力，缓解因特征稀疏导致的漏检问题；该模型引入动态检测头，该检测头通过动态注意力机制与统一尺度建模提升多尺度特征融合效果与小目标定位精度。结果显示：YOLOv5n-ChtaFSD模型的精确率、召回率与平均精度均值分别为90.7%、81.6%和85.9%，较原模型分别提升4.5百分点、4.7百分点和4.0百分点，明显优于YOLOv5n、YOLOv5s、YOLOv8n、YOLOv8s、YOLOv11n和YOLOv11s模型。YOLOv5n-ChtaFSD模型的参数量为2.4 M，浮点运算量为5.4 GFLOPs，模型体积为4.9 MB，表明该模型具有更优的检测性能，同时其参数量较小，有利于在移动机器人等边缘设备中部署应用。

Abstract:To address the problems of low efficiency and high cost in the current identification of Siam weed Chromolaena odorata, an improved object detection model YOLOv5n-ChtaFSD is constructed based on YOLOv5n. In the neck of the model, an efficient re-parameterizable generalized feature pyramid network(Efficient RepGFPN) is introduced, which not only improves detection accuracy but also reduces inference time by eliminating redundant upsampling operations. Meanwhile, the model embeds a simple parameter-free attention module(SimAM), which adaptively selects features based on feature importance, thereby enhancing the model's small-target perception capability and mitigating missed detection caused by sparse features. Furthermore, the model adopts a dynamic head(DyHead) that improves the effect of multi-scale feature fusion and small-target localization accuracy through dynamic attention mechanism and unified scale modeling. The results show that the precision, recall rate and mean average precision of the YOLOv5n-ChtaFSD model reach 90.7%, 81.6% and 85.9%, which are 4.5%, 4.7% and 4.0% higher than those of the original model, respectively, and the proposed model is significantly superior to YOLOv5n, YOLOv5s, YOLOv8n, YOLOv8s, YOLOv11n and YOLOv11s models. With 2.4 M parameters, 5.4 GFLOPs of computational complexity and a model size of 4.9 MB, YOLOv5n-ChtaFSD achieves superior detection performance in complex environments with fewer parameters, making it suitable for deployment on edge devices such as mobile robots.