摘要
白洋淀是华北平原最大的淡水湿地系统,素有“华北明珠”之称。本文结合无人机技术和目标检测算法以实现更好的对白洋淀内的生物进行监测和保护,为进一步评估物种的生存质量、确定保护等级以及制定相应保护策略提供科学依据。在YOLOv8n算法的基础上进行了改进,提出了DMW-YOLOv8算法。通过使用WIoU损失函数,更准确的衡量预测框和真实目标之间的相似性,提高多个物体时的检测精度;引入MLCA注意力机制,结合了通道信息和空间信息,以及局部信息和全局信息,以增强网络的表达能力;嵌入C2f_DCNv4算子优化稀疏DCN操作符以提高实际效率。实验结果显示,该算法的mAP50和mAP50-95分别达到89.4%和52.0%,相比原始YOLOv8n模型分别提高了2.7%和2.2%;每秒传输帧数达到177.8FPS相比原模型提升了71.2FPS。该算法在白洋淀生态环境的监测中具有较高的实用价值。
The Bai-yang Lake,known as the"Pearl of North China,"is the largest freshwater wetland system in the North China Plain.The unmanned aerial vehicle(UAV)technology coupled with object detection algorithms were used to enhance biological monitoring and conservation efforts in Baiyangdian Wetland,and to scientifically evaluate species'habitat quality,determine protection levels,and formulate corresponding conservation strategies,.ased on our comprehensive investigation,we have developed DMW-YOLOv8 as an advanced evolution of the YOLOv8n algorithm.By using the WIoU loss function,the similarity between predicted and actual targets is measured more accurately,improving detection accuracy,especially in multi-object scenarios.The MLCA attention mechanism is introduced,which integrates channel and spatial information,as well as local and global information,to enhance the network's representation ability.Additionally,the C2fDCNv4 operator is embedded to optimize the sparse DCN operator,improving practical efficiency.The proposed algorithm achieves superior performance metrics,with mAP50 and mAP50-95 reaching 89.4%and 52.0%respectively-representing improvements of 2.7%and 2.2%over the baseline YOLOv8n model.The system demonstrates remarkable computational efficiency,processing at 177.8 FPS(71.2 FPS faster than the original implementation).These advancements make our solution particularly valuable for ecological monitoring applications in the Baiyangdian wetland environment.
作者
彭凯远
田立勤
PENG Kaiyuan;TIAN Liqin(North China Institute of Science and Technology,Yanjiao 065201,China;School of Computer Science,Qinghai Normal University,Xining 810016,China)
出处
《华北科技学院学报》
2025年第3期56-66,共11页
Journal of North China Institute of Science and Technology
基金
河北省重点研发计划项目(H52627482-19270318D/01)
廊坊市科技支撑计划项目(2024011086)
中央高校基本科研业务费项目(050201080303)。
