基于LVT-YOLOv8的草莓品质检测方法OA

To address the issues of low efficiency and limited feature extraction in traditional visual strawberry quality detection methods,this paper proposes an improved YOLOv8s network model for strawberry appearance quality detection.Based on the YOLOv8s baseline network,we introduce the LVT-YOLOv8 enhanced model with the following improvements.Firstly,the back-bone network is optimized into a lightweight Vision Transformer model,RepViT m1,to better capture the spatial information in images.Secondly,by integrating StarNet in the neck network to construct the C2f-Star module,the architectural complexity of the Neck network can be significantly simplified.This not only reduces the computational load and parameters of the model but also enhances the performance of feature fusion by expanding the dimension of the feature space.Thirdly,a lightweight shared convolutional BN(Batch Normalization)detection head is proposed to maintain the detection capability that is sensitive to multi-scale feature information.Finally,an improved Inner-WIoU loss function is adopted to enhance the regression loss of bounding boxes.The improvement in parameters and size reduced by 6.4%and 4.7%respectively,while the computational load only in-creased by 6.0%.The detection speed FPS reached 163.9 fps,and the precision,recall rate and mean average precision are re-spectively increased by 2.2 percentage points,2 percentage points and 1.6 percentage points compared to the original model,reaching 98.6%,98%and 95.3%.In conclusion,the improved LVT-YOLOv8 model can meet the algorithm requirements for automated production in the strawberry industry and can provide technical support for strawberry automated sorting.