华中科技大学学报（自然科学版）2024，Vol.52Issue(3)：35-40,46,7.DOI:10.13245/j.hust.239405
基于MCA-YOLO的轻量级红外实时目标检测算法
Lightweight infrared target real-time detection algorithm based on MCA-YOLO
摘要
Abstract
Infrared object detection in complex ground scenes often faces challenges such as low detection accuracy and the difficulty of deploying large network models on mobile or embedded platforms.To address these issues,a lightweight real-time infrared object detection model called MCA-YOLO was proposed.Based on the YOLOv4 model,the backbone network adopted the Mobilenet-v2 architecture,while the remaining parts of the network employ depth-wise separable convolutions to replace standard convolutions,reducing the model parameters and computational complexity.In the neck network of the model,a coordinate attention(CA)module was embedded to enhance the model's feature extraction capabilities.The model Anchor was redesigned using the k-means clustering algorithm to improve detection accuracy.Additionally,a transfer learning strategy was employed for pretraining the model to accelerate convergence.Experimental results demonstrate that the proposed detection model achieves an 81％reduction in model parameters,a 47％increase in detection speed,and a 6.34％improvement in detection accuracy compared to YOLOv4.The model ensures both accuracy and real-time performance,thereby meeting the requirements for real-time infrared object detection in military and security fields.关键词
红外图像/目标检测/注意力机制/轻量级/迁移学习Key words
infrared image/target detection/visual attention/lightweight/transfer learning分类
信息技术与安全科学引用本文复制引用
刘冬,李庭鑫,杜宇,丛明..基于MCA-YOLO的轻量级红外实时目标检测算法[J].华中科技大学学报（自然科学版）,2024,52(3):35-40,46,7.基金项目
装备预研教育部联合基金资助项目(8091B022119). （8091B022119）
