摘要
国家珍稀动物丹顶鹤多栖息于开阔平原、沼泽、湖泊、海滩及近水滩涂等地,不利于人工跟踪及近距离观测丹顶鹤的实时动向。针对传统快速卷积神经网络(Faster RCNN)检测不准确的问题,现设计提出在原始网络基础上,结合多特征融合金字塔(FPN)实现多层语义信息融合,建立针对丹顶鹤的识别模型。此外,为提高模型召回率,还采用了线性加权的非极大值抑制算法。改进后的算法较原始算法精确率提高了3.5%左右。实验结果表明,所提模型对于丹顶鹤种群具有较好的识别效果,为湿地的丹顶鹤保护与监测提供新的技术手段,使湿地的生态文明建设得到进一步加强。
Red-crowned cranes,a rare national animal,mostly inhabit in open plains,swamps,lakes,beaches and near tidal flats,which is not conducive to manual tracking and real-time observation of red-crowned cranes.Aiming at the inaccurate detection problem of traditional Faster RCNN,the present design proposes to realize multi-layer semantic information fusion based on the original network and combine FPN to establish the identification model for red-crowned cranes.In addition,in order to improve the recall rate of the model,a linear non-maximum suppression algorithm is used.The accuracy of the improved algorithm is about 3.5%higher than that of the original algorithm.The experiment result shows that the model has a good identification effect for red-crowned cranes,it provides a new technical means for the protection and monitoring of red-crowned cranes in wetland,and further strengthens the ecological civilization construction of wetland.
作者
吕秀丽
陈帅男
Lyu Xiuli;Chen Shuainan(School of Physics and Electronic Engineering,Northeast Petroleum University,Daqing 163318,China)
出处
《电子测量技术》
2020年第20期104-108,共5页
Electronic Measurement Technology
基金
黑龙江省高等教育教学改革研究项目(SJGY20180069)资助。
