基于改进ResNet的机场鸟类识别方法  

Airport bird recognition method based on improved ResNet

作　　者：孔建国[1] 赵志伟 张向伟 梁海军[1] KONG Jianguo;ZHAO Zhiwei;ZHANG Xiangwei;LIANG Haijun(College of Air Traffic Management,Civil Aviation Flight University of China,Guanghan 618307,China)

机构地区：[1]中国民用航空飞行学院空中交通管理学院,四川广汉618307

出　　处：《电子设计工程》2025年第5期172-177,共6页Electronic Design Engineering

基　　金：中央高校基本科研业务费专项资金资助(PHD2023-035,ZHMH2022-009)。

摘　　要：针对机场鸟类识别过程中存在识别难度较大、准确率较低等问题,该文提出了一种改进ResNet的SA-ResNet(SPDConv and Attention-ResNet)模型。模型采用空间到深度卷积(SPDConv)替换ResNet18中的跨步卷积层,避免信息的过度丢失,增强模型特征提取能力;使用高效通道注意力(ECA)改进卷积块注意力模块(CBAM),并提出高效卷积块注意力模块(ECBAM)进一步提高模型识别准确率。通过自建的ADB-20机场鸟类数据集验证表明,SA-ResNet模型的准确率达到了95.9%,能够很好地识别机场鸟类,为机场开展鸟击防范工作奠定基础。An improved ResNet model,SA-ResNet(SPDConv and Attention-ResNet),is proposed to address the challenges of difficulty and low accuracy in airport bird recognition.The model replaces the strided convolution layers in ResNet18 with Space-to-Depth Convolution(SPDConv)to avoid excessive information loss and enhance the model’s feature extraction capabilities.Additionally,the model incorporates Efficient Channel Attention(ECA)to enhance the Convolutional Block Attention Module(CBAM)and introduces the Efficient Convolutional Block Attention Module(ECBAM)to improve recognition accuracy further.Validation on the self-built ADB-20 airport bird dataset demonstrates that the SA-ResNet model achieves an accuracy of 95.9%,effectively identifying airport birds and laying the foundation for bird strike prevention at airports.

关 键 词：鸟类识别 残差网络 注意力机制 深度学习 

分 类 号：TN919.8[电子电信—通信与信息系统]