基于FPB-DETR的苹果成熟度检测算法

打开文本图片集

中图分类号：TP391.41 文献标识码：A DOI：10.7535/hbkd.2026yx02010

Abstract：Toaddress the lowacuracy and eficiencyof apple maturitydetection under large-scale，lighting，andoclusion conditions，animprovedFPB-DETRdetection model basedonRT-DETRwasproposed.Firstly，afrequency-adaptivedilated convolution（FADC）module wasintroduced into the backbone network to precisely focusonsubtlecolor gradients，immature spots，adtexture stripes onapplesurfaces byresolving theconflictbetween efective receptivefieldandfeature bandwidth，as wellasvercomingthelimitationsoffixeddilationrates.Secondly，apolaformeratention-basedintra-scalefeatureinteraction （Pola-AIFI） module was designed to mitigate the issues of negative value neglect andexcesive information etropy， suppressinginterference fromtargetapplesundervarying environmentalconditions.Finall，abi-directionalfeature pyramid network（BIFPN）structure was introduced during the multi-scale fusion stage tooptimizefeature fusion eficiency and key information focusingcapability，reducingambiguity interference inmaturityfeature transmision.Theresultsshowthatthe precision，recal rate and average accuracy of the FPB-DETR model proposed in this study are 92.5% ， 92.7% and 96.8% ， respectively，which increases by 2.0% ， 1.7% and 1.8% ，respectively compared with the original model，and are superior to those of Faster R-CNN，YOLOv5m， YOLOv8m ，YOLOvllm and YOLOvl2m object detection models， significantly enhancing the detection capability of the model; The average detection time of the model is 31ms ，which meets the real-time detectionrequirements forapple maturity.This studyrealizes beterdetectioneffectbycombiningfeatureextraction，attntion mechanism and multi-scale fusion，providing reference for the optimization design of intellgent harvesting robots.

Keywords： computer neural networks; object detection; maturity；multi-scale fusion;RT-DETR

中国作为全球最大的苹果生产国，2024年年产量高达3734万t，然而出口量仅为98万t，不足总产量的 3% ，这一数据与中国苹果生产大国的地位极不匹配[1]。（剩余14422字）