基于改进FasterR一CNN的水稻秧苗漏插识别研究

打开文本图片集

中图分类号：S223.91；TP391.4 文献标识码：A 文章编号：2095-5553（2026）02-0101-07

Research on rice seedling omission identification based on improved Faster R一CNN

Zou Liwen，Liang Chunying，Hu Jun，Chen Yuheng，Li Zhenpeng （College of Engineering，Heilongjiang Bayi Agricultural University，Daqing，163319， China）

Abstract：Rice，as the main graincropin China，achieving high yieldand excellent output isan inevitable trend．In response to theproblems of sloweficiencyand high subjectivityin traditional manual seedlingsuplementation，arice seedingomissionidentification method basedonimprovedFasterR—CNNisproposed.Basedon theFasterR—CNN model，thebackbonenetwork isreplacedwiththeresidualnetwork ResNet5O，andfeatureinformationisextractedin combinationwith theFPN featurepyramid.TheideaofRoIAlignbilinear interpolation isintroduced toreplace the operationofroughquantization inRoIPooling layers.The experimentalresultsshow that theacuracy rateof recognition of the improved Faster R—CNN model is 93.62% and the mean average accuracy in recognition mAP@0.5 is 95.06% ： Compared with the unimproved model，the recognition accuracy is increased by 7.33% ，and the average accuracy of the model mAP@0.5 is increased by 4.6% . The results indicate that this study can improve the classification of rice seedlings and the detection of missd planting positions in rice transplanters，laying a solid foundation for developing riceseedling replacement plans and providing data support for evaluating the quality of rice transplanters.

Keywords：rice seedlings；omission identification； feature pyramid；deep learning；residual network

0 引言

我国作为水稻的生产和出口大国，超过 60% 的人以稻米为主食[1]。（剩余10905字）