指挥控制与仿真2025，Vol.47Issue(5)：42-48,7.DOI:10.3969/j.issn.1673-3819.2025.05.006
面向高价值目标识别的SKConv-MobileNetV3改进
SKConv-MobileNetV3 improvements for high-value target recognition
摘要
Abstract
High-value target recognition usually requires high timeliness and accuracy,and the large number of parameters of traditional deep convolutional neural networks leads to a large number of application scenarios being limited.A high-value target recognition method based on SKConv-MobileNetV3 is proposed,by fusing the recognition results of SKConv convolu-tional kernel with the weighted output feature maps of MobileNetV3 convolutional layer,and using the attention mechanism to select the most relevant image content for feature extraction,we can improve the feature information extraction capability of the SKCon-MobileNetV3 model while improving learning efficiency,under the condition that the number of parameters re-mains unchanged,MobileNetV3 model's ability to extract feature information,while improving the learning efficiency and recognition accuracy.Through the above algorithm improvement and optimization,seven high-value targets with strong corre-lation are selected in the NWPU-RESISC45 dataset and tested,and the results show that the accuracy is improved by 7.01%compared with MobileNetV2 and 4.08%compared with MobileNetV3,which is able to better improve the recognition accura-cy of high-value target recognition accuracy.关键词
计算机应用/目标识别/轻量级网络/MobileNetV3/自适应选择性卷积Key words
computer application/target identification/lightweight network/MobileNetV3/selective kernel convolution分类
信息技术与安全科学引用本文复制引用
郑鹏,程云,刘波,叶晨浩,王石杰..面向高价值目标识别的SKConv-MobileNetV3改进[J].指挥控制与仿真,2025,47(5):42-48,7.基金项目
国家社科基金重点项目(2023-SKJJ-B-107) （2023-SKJJ-B-107）
