Research on Oranges Target Recognition Based on Improved YOLOv5
Aiming at the problems of low accuracy of existing oranges target recognition,as well as the large amount of parameters and floating-point calculations of deep learning models,this paper proposes to improve the YOLOv5 algorithm in three aspects,one is to introduce lightweight networks Mobilenetv3,Shufflenet V2,Ghost,etc.to improve the Backbone module of YOLOv5,the second is to improve the attention mechanism of the C3 part of the neck network,and the third is to use Ghost Conv module to improve the Conv module of the NECK network.Finally,the improved algorithm reduces the amount of model parameters and floating-point computation to about 1/7 of the original,and after training on optimized parameters,the model reaches 0.957 mAP@0.5 on the test dataset.
NETL
NSTL
万方数据
维普
