Aiming at the problem of inaccurate strawberry identification caused by complex environments such as light,occlusion,dense fruits and uneven distribution,this paper proposes an improved model based on YOLOv7 to YOLOv7-SCC to create a strawberry sample dataset to establish real picking complex environment data.First,use ShuffleNetv2 to replace the YOLOv7 backbone network to achieve lightweight and effectively reduce the amount of model parameters;Secondly,the CBAM attention mechanism module is introduced to enhance the recognition of strawberry areas by the feature network;finally,Content-Aware ReAssembly of FEatures(CARAFE)upsampling is selected to expand the receptive fields in the feature fusion network and make full use of semantic information.After experiments,the parameter quantity of the improved model is reduced by 59%,the floating point number is reduced by 68.2%,and the accuracy rate is 99.6%.The results proved that the improved YOLOv7-SCC can accurately identify strawberry ripeness while maintaining high accuracy,making it a more suitable choice for high-ridge strawberry ripeness detection than other algorithms.
万方数据
维普
