| 摘 要: 针对野生动物目标检测过程中存在的实时性不足、复杂背景干扰以及多尺度目标检测难度较大的问题,提出了一种基于改进 YOLOv11的野生动物目标检测模型。首先,设计了用 DCA-C3k2模块替代 C3k2模块,并且在 MSA-C3k2模块中引入 CBAM 和高效通道注意力(ECA)机制,增强模型对关键特征的聚焦能力;同时采用多尺度动态卷积核,通过注意力权重动态融合特征;其次,用 MSA-SPPF模块替代SPPF模块,在原SPPF模块基础上添加了多尺度卷积块,并且引入了多头自注意力机制,增强了SPPF模块的特征表达能力;最后,提出一个重构的检测头模块 DSED-Detect,使用动态卷积自适应提取特征,使用SEBlock增强通道注意力,提升边界框回归精度,轻量化 GhostConv减少参数量,使用混合注意力增强类别判别力。实验结果表明,与原 YOLOv11模型相比,改进后的模型 mAP50提高了4个百分点,mAP50∶95提高了4.6个百分点,精确率提高了7.6个百分点,由此证明改进后的模型能够更好地运用于野生动物的目标检测任务。 |
| 关键词: 深度学习 目标检测 YOLOv11 注意力机制 野生动物 |
|
中图分类号: TP391
文献标识码: A
|
|
| Wild Animal Object Detection Model Based on Improved YOLOv11 |
|
YAN Weiqiang, LIU Liqun
|
(College of Information Science and Technology, Gansu Agricultural University, Lanzhou 730070, China)
yanwq@st.gsau.edu.cn; llqhjy@126.com
|
| Abstract: To address the challenges of insufficient rea-l time performance, complex background interference, and difficulties in mult-i scale object detection in wild animal monitoring, a wild animal object detection model based on an improved YOLOv11 is proposed. First, the DCA-C3k2 module is designed to replace the C3k2 module. The MSA-C3k2 module incorporates CBAM and ECA attention mechanisms to enhance the model’s ability to focus on key features,while a mult-i scale dynamic convolution kernel is introduced to dynamically fuse features through attention weights.Then, the MSA-SPPF module is used to replace the SPPF module. Based on the original SPPF module, a mult-i scale convolutional block is added, and a mult-i head sel-f attention mechanism is introduced to enhance the feature representation capability of the SPPF module. Finally, a reconstructed detection head module, DSED-Detect, is
proposed. It employs dynamic convolution for adaptive feature extraction, SEBlock to enhance channel attention and improve bounding box regression accuracy, lightweight GhostConv to reduce parameter count, and hybrid attention to enhance class discriminability. Experimental results show that compared with the original YOLOv11 model, the improved model achieves a 4 percentage points increase in mAP50, a 4.6 percentage points increase in mAP50∶95,and a 7.6 percentage points increase in precision, demonstrating that the improved model is better suited for wild animal object detection tasks. |
| Keywords: deep learning object detection YOLOv11 attention mechanism wild animal |
