Research on Supermarket Abnormal Behavior Detection for Edge End
This paper proposes a lightweight YOLOv5 network model for abnormal human behavior detection in super-markets.This method uses Ghost convolution to replace the standard convolution to reconstruct the network,reduces the number of model parameters and calculation to adapt to low computing power devices at the edge,and adds a decoupled fully connected attention mechanism that is more conducive to the application of mobile terminal architecture devices to capture more long-distance spatial feature information and improve the detection accuracy.Experimental results show that compared with the classic YOLOv5 detection method,the accuracy of this method is increased by 4.6%,the calculation amount is reduced by 43%,the parameter amount is reduced by 42%,and the detection speed is increased by 67%.The detection FPS on the edge device Sunrise x3 can reach 17.82.
NETL
NSTL
万方数据
