PY一YOLO：基于深度学习的梨幼果目标检测方法

打开文本图片集

中图分类号：S126 文献标识码：A 文章编号：2095-5553（2026）03-0097-06

DOI：10.13733/j.jcam.issn.2095-5553.2026.03.014

Abstract：Addressingtheissuesofhighlaborintensityandlowinteligencelevelinmanualdetectionofyoungpearfruits，a object detection method foryoung pear fruitsbased ondeplearning named PY—YOLOis proposed.This modelinnovatively employsthe DRBNCSPELAN module in both the detection backboneanddetectionheadforfeatureextraction， integratigamulti-branch structureandreparameterization techniques toenhancethemodel's featurelearningabilityandre parameterizationeficiency.Thissignificantlyimproves therecognitionandlocalizationaccuracyforyoungpearfruits. Meanwhile，theEMAattentionmechanismisintroduced，andcros-spatial learningmethodsareadoptedtofurther efectivelycaptureandutilizemulti-scalefeatureinformation，therebyfurtherstrengtheningtherecognitioncapability and robustnessof the young pear fruit detection model.Through comparative analysis of diferent loss functions，the studyconfirms thesuperiorityof the SIoUlossfunction inenhancinglocalizationaccuracyand modelrobustness inthecontext of young pear fruit detection.On the test set，the PY—YOLO model exhibitsoutstanding performance，achieving notable improvements over the YOLOv8 model，with a precision of 90.57% ，a recall rate of 87.74% ，an mAP of （204号 93.26% ，and maintaining a framesper second（FPS）rate of 62 frame/s，thereby fulfillng the needforreal-time and accuratedetectionofyoung pear fruitsinactual pear orchards.Consequently，thisstudy offersaneficientand accurate method for young pear fruit detection，showcasing itsbroad application potential in agricultural automation.

Keywords：young pear fruits；deep learning；YOLO；object detection； feature extraction

0 引言

试验与方法

在梨园生产管理中其幼果期的有效管理对于最终果实的品质和产量具有决定性的影响。（剩余9349字）