基于DETR网络的棉花顶芽精准识别方法

打开文本图片集

中图分类号：S562；TP391.4 文献标识码：A 文章编号：2095-5553（2026）02-0108-06

Abstract：Toaddress theisuesof low detection accuracyand poor robustness incottnapical budrecognition under complex environments and varying growth stages，thispaper proposes an enhancedcotton apicalbudrecognition model based onthe Detection Transformer（DETR）framework.The proposed model employs Convolutional Neural Network （CNN）technology to extract discriminative features fromcottnapical bud imagesandthe Transformerarchitecture to modelcontextualrelationshipsamong budmorphologies incomplex backgrounds，efectivelymitigatingfeature degradation．Byintroducingabipartitematchingloss functioncombinedwithend-to-endoptimization，the framework achieves enhanced recognition precision.Furthermore，we integrate Deformable Convolutional Networksv2（DCNv2）into thebackbonenetworktoimprovethefeatureextractioncapabilityforapicalbudcharacteristics.Experimentalresults demonstrate that the improved model achieves a mean Average Precision （ .mAP ）ofO.83and ameanaccuracy of0.85 underheterogeneousilluminationconditions，validatingitsrobustnessagainstperturbationsinducedbybothlighting varitionsanddevelopmental stage diferences.Thisapproach providesanefective technical solutionfortheprecise identificationofcotton apical budsacross growth phases incomplex field environments，laying acomputational foundation for intellgent cotton topping operations.

Keywords：cottnapical buds；detection transformer（DETR）；convolutional neural network（CNN）；deep learning; loss function

0 引言

棉花顶芽作为棉花植株生长发育的关键部位，不仅决定了植株的整体形态和分枝结构，还对光合产物的分配和最终产量具有重要影响[12]。（剩余9244字）