复杂场景下ResNet34优化算法在苹果检测中的应用研究

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）04-0133-06

Abstract：To improve the detectionaccuracyof apple fruits incomplex environments and reduce the missd detection rate，theResNet34modelisstructurallyoptimizedand improved.TheSEatention mechanism is introduced indiferent feature extraction layersof this model，enabling thenetwork tofocusonthepositionof aple fruitsandefectively cope withtheinterference of complex backgrounds.Aimingattheproblemssuchas thelong distanceof apple fruits，smal targetsand severe leaf occlusion，the Res2Net network module isadopted.Thismodulehas theabilityto extract multi-scale informationandcantakeintoaccounttargetsof diferentsizes intheimage.Furthermore，thedualdetection headstrategy isadoptedtoimprovethemodel，furtherenhancing thecomprehensiveperformanceofthemodel indetecting smalltargets.The results show that the detection precision rate of the improved model on the test set is 83.1% ，the F1 score is 77.3% ，and the mean average accuracy mAP is 82.3% ，which are increased by 3.7% ， 2.2% and 4% （20 respectively compared with the original ResNet34 model.

Keywords：apple；automated picking；complex scenarios；SE attention mechanism；object detection

0 引言

目前，苹果的采摘仍然以人力为主，不仅采摘速度慢，也增加了人力和物力的成本。（剩余9536字）