摘要
【目的】针对传统鸟鸣声识别方法中特征提取不全面等问题,提出一种基于注意力机制的鸟声识别模型,以期提高鸟声识别精度,完善鸟类监测体系。【方法】对ResNet34模型进行卷积核优化,引入注意力机制——CBAM模块,并对CBAM模块结构进行优化,构建CRNet识别模型。利用xeno-canto网络数据库构建的神农架珍稀鸟类数据集对模型进行验证分析,以实现神农架国家公园鸟类的声音识别。【结果】1)在模型精度方面,所提出的鸟类声音识别模型对鸟类声音识别的准确率可达89%,相比CBAM-ResNet和ResNet34模型,准确率分别提升了2%与9%。模型在大多数类别上的预测效果良好,准确率在85%以上。2)在模型参数方面,CRNet的参数量为5.36 M,计算量为73.33 M,相比ResNet34的参数量降低了约75%,同时计算量相比ResNet34降低了约3%。3)与其它深度学习模型进行了对比,表明CRNet在F1分数、mAP、准确率的结果均优于MelResNet和VGG11。【结论】提出的神农架鸟类声音识别模型具有较高的识别准确率,为神农架国家公园的鸟类保护与监测提供了技术支撑。
【Objective】A bird voice recognition model based on attention mechanism is proposed to address the issues of incomplete feature extraction in traditional bird voice recognition methods,in order to improve the accuracy of bird voice recognition and improve the bird monitoring system.【Method】We optimized the convolution kernel of the ResNet34 model,introduced the attention mechanism—CBAM module,and optimized the structure of the CBAM module to construct a CRNet recognition model.We used the xeno-canto network database to construct a dataset of rare birds in Shennongjia to validate and analyze the model,in order to achieve sound recognition of birds in the Shennongjia region.【Result】1)Regarding model accuracy,the proposed bird sound recognition model achieves an accuracy of 89%for bird sound recognition,which is 2%and 9%higher than the CBAM-ResNet and ResNet34 models,respectively.The model has good prediction performance in most categories,with an accuracy of over 85%.2)Regarding model parameters,CRNet has a parameter count of 5.36 M and a computational load of 73.33 M,which is about 75%lower than ResNet34 in terms of parameter count,while the computational load is about 3%lower than that of ResNet34.3)Compared with other deep learning models,CRNet outperforms MelResNet and VGG11 regarding F1 score,mAP,and accuracy.【Conclusion】The proposed Shennongjia bird sound recognition model has a high recognition accuracy,providing technical support for the protection and monitoring of birds in the Shennongjia area.
作者
刘昱坤
邓广
于新文
陈艳
郭安琪
杨蔡芸
李奕阳
Liu Yukun;Deng Guang;Yu Xinwen;Chen Yan;Guo Anqi;Yang Caiyun;Li Yiyang(Institute of Forest Resource Information Techniques,Chinese Academy of Forestry,Beijing 100091,China;Key Laboratory of Forestry Remote Sensing and Information System,NFGA,Beijing 100091,China)
出处
《陆地生态系统与保护学报》
2024年第3期39-48,共10页
Terrestrial Ecosystem and Conservation
基金
中国林业科学研究院基本科研业务费专项资助(CAFYBB2023ZA005-2)。
关键词
鸟类监测
鸟声识别
注意力机制
bird monitoring
bird sound recognition
attention mechanism
