Research On Improving The Accuracy Of YOLOv5 Object Detection Using Attention Mechanism

In recent years,the wave of artificial intelligence triggered by deep learning has swept across various fields.Deep learning uses neural networks to model data and deep networks to learn complex patterns.It has been widely used in computer vision,natural language processing,speech recognition and other fields.Object detection is an important branch of computer vision technology.Its purpose is to identify objects of interest from images or videos,and determine their positions.It is applied to autonomous vehicle,unmanned aerial vehicles technology,security monitoring and other fields.YOLOv5 is the fifth generation product of the YOLO(You Only Look Once)series of object detection algorithms.It is currently one of the most popular object detection algorithms,but there is still a problem of insufficient feature extraction ability.This article proposes a fusion scheme that inserts attention mechanism into the backbone network of YOLOv5 network model to address the lack of effective feature enhancement and background noise suppression in YOLOv5.Based on YOLOv5 s,four typical attention mechanisms were selected for modular design,namely SE(Squeeze and Excitation),CBAM(Convolutional Block Attention Module),ECA(Efficient Channel Attention),and CA(Coordinated Attention),and inserted into the YOLOv5 s network model to improve it.A network model with a single position insertion attention mechanism was constructed to study the performance impact of insertion position on the overall network model of YOLOv5 s.Subsequently,a YOLOv5s＿Attention network model integrating attention mechanism was further constructed to enhance YOLOv5s’ feature extraction ability,suppress background noise,and achieve the effect of improving its target detection accuracy.This paper uses Microsoft’s large public dataset,MS COCO,to train and validate four network models that insert attention mechanism into a single location and YOLOv5s＿Attention network models that fuse attention mechanism.The performance of these models is evaluated using COCO evaluation index.After comprehensive testing and analysis,the evaluation index of the YOLOv5s＿Attention network model which fuses attention mechanism is better than YOLOv5 s,and the quantity of parameters and calculation changes after improvement.Compared with the original network model,the YOLOv5s＿Attention network model incorporating CA attention mechanism has an AP(Average Precision)increase of 0.8 percentage points.AP@0.50 Increased by 1.2 percentage points,AP@0.75 Increased by 0.6 percentage points.