基于残差卷积神经网络模型的勺嘴鹬动作识别  

Action recognition of spoon-billed sandpipers(Eurynorhynchus pygmeus)based on residual convolutional neural network model

作　　者：杨雪珂 蒙金超 冯悦恒 林婷婷[1,2] 王兆君 刘辉 YANG Xueke;MENG Jinchao;FENG Yueheng;LIN Tingting;WANG Zhaojun;LIU Hui(School of Forestry,Hainan University,Haikou,Hainan 570228,China;Institute of Zoology,Chinese Academy of Sciences,Beijing 100101,China)

机构地区：[1]海南大学林学院,海口570228 [2]中国科学院动物研究所,北京100101

出　　处：《热带生物学报》2023年第5期481-489,共9页Journal of Tropical Biology

摘　　要：为开启海南热带地区鸻鹬类涉禽的动作识别以及其他野生鸟类行为学自动识别的研究,建立了基于野外采集影像的勺嘴鹬(Eurynorhynchus pygmeus)动作图像数据集。该数据集由表达勺嘴鹬主要行为模式的9种动作标签组成;同时利用ResNet50、ResNet101和ResNet152共3种残差卷积神经网络模型尝试对勺嘴鹬的动作进行自动识别。结果表明,ResNet50、ResNet101、ResNet152测试集准确率分别为96.90%、96.94%和96.90%,说明3种模型都能对勺嘴鹬图像进行快速准确的动作识别。With the widespread application of image acquisition equipment and data sharing platform,the amount of bird image data has been increasing at an unprecedented speed.How to effectively deal with such a large amount of data has become a major challenge.In recent years,convolutional neural network has shown strong practicability and effectiveness in the application of automatic bird image processing.However,there has been no research on automatic recognition of movements in wild birds.In view of this,a special action image dataset of the sandpiper was established based on field images.The dataset was composed of nine action tags representing the main behavior patterns of spoon-billed sandpipers(Eurynorhynchus pygmeus).At the same time,three residual convolutional neural network models,ResNet50,ResNetT101 and ResNet152,were used to automatically recognize the movements of the spoon-billed sandpipers.The experimental results showed that the three models achieved excellent results in action recognition with their accuracy rates of the test set being 96.90%(ResNet50),96.94%(ResNet101)and 96.90%(ResNet152),respectively.This indicates that these three models have a rapid recognition of the movements of the spoon-billed sandpiper.

关 键 词：残差卷积神经网络 鸟类图像 动作识别 勺嘴鹬 

分 类 号：Q958[生物学—动物学]