基于改进YOLOv11的监控视频小目标检测算法OA

In order to solve the problems of weak feature expression,high miss rate,large background interference and inaccurate re-gression of bounding box caused by image deformation in small target detection at long distance and high viewing angle in surveillance video,an improved target detection algorithm BEH-YOLOv11n is proposed.Firstly,the structure of C3K2 is replaced by MSBlock module to enhance the model's perception of targets of different scales.Secondly,HS-FPN network is introduced to replace the original neck net-work to improve the expression ability of small targets in high and low level feature maps;Finally,the attention mechanism of EMA is intro-duced into C2PSA,and the regression loss function of bounding box is replaced by CIoU,which enhances the modeling ability of the model for the spatial position and direction of small targets.The experimental results based on VisDrone2019 data set show that the accuracy of the improved algorithm is increased by 9 percentage points,the mAP50%is increased by 5.6%compared with the original algorithm,and the detection speed FPS is 112 frames/s.It meets the standards of real-time detection while maintaining accuracy and accuracy,and meets the requirements of small target detection in complex monitoring environment.