基于改进YOLOv8的瓶装产品褶皱检测模型

打开文本图片集

Wrinkle Detection Model for Bottled Products Based on Improved YOLOv8

CAO Haiwen1，3CHEN Deping3，4YANG Danni³ WANG Nan³ ZHONG Zhenyu³DUAN Erqiang2 （1.Guangdong University of Technology， Guangzhou 510o06， China 2.Foshan VIOMI Electrical Technology Co.， Ltd.，Foshan 528308， China 3.Institute of Intelligent Manufacturing， Guangdong Academy of Sciences， Guangzhou 510o70， China 4.Shantou University，Shantou 515063，China）

Abstract：Aimingattherandomrotationangles，multi-scalecharacteristics，andlowcontrast withthe botle backgroundof wrinkles nottedproducts，thispapepropossanimprovedYOLOv-basedwkleetectionmodelforbottedproducts.Tedel utilizes the Sobel operator to extractedge features and enances edge extraction capabiltyinlow-contrastregions byfusing edge features withspatialfeatures.Byintegratingconvolutionandadditiveatentionmechanisms，theYLOv8modelcansimultaneously extractglobalandlocalfeaturesoftheimage，improving itsabilitytofeaturefusion wrinklesofdifferentscalesandenhancing robustnessagainstrotated targets.Thfastspatialpyramid poling moduleiseplacedwithanintra-scalefeatureinteractionmodule based onatention，strengthening the backbone network'sabilityto extract wrinkle features within the same scale anddecoupling featureextractionfrommulti-calefeaturefusiontasks.Thepathagregationnetworkisimprovedusingabidirectionalfaturepyamid network，whichdynamicalyadjusts feature weights through weightedfeaturefusion toenhancethe model'sfeaturefusioncapability. Basedonthis model，a wrinkledetectionsystemforbotled productsisconstructed.Experimentalresultsshowthat theimproved YOLOv8 model achieves an AP50 of 82.3% a 6.6% improvement over the YOLOv8-OBB model， demonstrating superior performance in wrinkle detectionforbotled products.Verifiedonactual productionlines，thesystem mets industrialrequirementsintemsof detection accuracy，inference speed，and stability.

Keywords：botledproducts wrinkledetection;improved YOLOv8;Sobeloperator;aditiveatentionmechanism;attentionbased intra-scale feature interaction module; bidirectional feature pyramid network

0 引言

瓶装产品褶皱检测是指通过视觉检测技术对瓶体外观进行检测并识别表面褶皱，以确保包装质量完整，避免因褶皱导致内容物泄漏或污染。（剩余12170字）