复杂背景下红外图像弱小目标检测

红外弱小目标检测是安防监控、侦察探测、精确制导等领域的关键技术。为了提高复杂背景条件下红外弱小目标检测的准确性和实时性，提出了一种基于深度学习的红外弱小目标检测算法YOLO-FCSP。根据红外图像中弱小目标的特点，在YOLO检测框架的基础上，通过减少下采样次数，结合跨阶段局部模块、Focus结构和空间金字塔池化结构设计了特征提取网络。借鉴多路径聚合的思路优化特征融合网络，同时调整检测输出层数量，通过信息复用提高特征利用效率。实验结果表明，本文提出的算法在检测红外弱小目标时具有较高的准确率和检测速度，精度和召回率分别为91.9%和94.6%，平均准确率(AP)值达到92.6%，检测速度达到170 f/s，满足实际应用中实时检测的需求。

Dim-small targets detection of infrared images in complex backgrounds

The infrared dim-small target detection is a key technology in the fields of security surveillance, reconnaissance detection and precision guidance. To improve the accuracy and real-time performance of infrared dim-small target detection under complex background conditions, an infrared dim-small target detection algorithm YOLO-FCSP based on deep learning was proposed. Considering the characteristics of dim-small targets in infrared images, the feature extraction network was designed based on the YOLO detection framework by reducing the number of downsampling and combining cross-stage local modules, Focus and spatial pyramidal pooling structure. The feature fusion network was improved by utilizing the idea of multi-path aggregation. The number of detection output layers was adjusted to enhance the reuse of feature information. The experimental results show that the proposed method has higher accuracy and detection speed in detecting infrared dim-small targets, achieving 91.9% precision and 94.6% recall, the average precision (AP) value to 92.6%, and the detection speed as 170 fps, which meet the requirements of real-time detection in practical applications.