基于改进YOLO v8模型的轻量化梨果实检测OA

To address the challenges of poor detection performance of target detection models for pear-picking robots in complex orchard environments and the difficulty of deployment on resource-constrained embedded platforms,a lightweight pear fruit detection model based on an improved YOLO v8 is proposed in this study.First,a novel PR_Bottleneck module was intro-duced,which employed an innovative structural design to effectively reduce the model's parameter count and computational load,significantly easing the operational burden.Second,the SPPF_CBAM module was proposed to greatly enhance the model's focus on target regions,thereby significantly improving detection accuracy in complex scenarios.The model also incorporated an ADown downsampling module,which not only reduced model complexity and parameter count but also strengthened information retention during the downsampling process.Finally,the WIoU loss function was introduced,which enabled the model to concentrate on medium-quality anchor boxes during training,thus improving overall detection performance and enhancing the model's generali-zation capability.Experimental results showed that compared with the original YOLO v8 model,the precision,recall,and mean average precision(mAP50)of the improved model increased by 0.7,3.7,and 1.7 percentage points,respectively.The parameter count and computational load were reduced by 38.72％and 30.86％,respectively.The model size was reduced by 38.33％,and the detection speed reached 85.4 frames per second,an improvement of 12.8 frames per second over the original model.Therefore,the improved YOLO v8 lightweight model not only improves detection accuracy and speed,but also sig-nificantly decreases the computational load and parameter count,enabling rapid and effective real-time pear fruit detec-tion on embedded platforms.