基于改进Mask R-CNN的马铃薯芽眼识别方法  

Potato Bud Eyes Recognition Method Based on Improved Mask R-CNN

作　　者：吴海风 黄操军[1] Wu Haifeng;Huang Caojun(College of Information and Electrical Engineering,Heilongjiang Bayi Agricultural University,Daqing 163319)

机构地区：[1]黑龙江八一农垦大学信息与电气工程学院,大庆163319

出　　处：《黑龙江八一农垦大学学报》2024年第4期98-105,共8页journal of heilongjiang bayi agricultural university

摘　　要：马铃薯芽眼的准确识别是马铃薯种薯自动化切块的前提,为了提高马铃薯芽眼的识别效果,解决长期以来人工切块效率低下问题,提出了一种基于改进Mask R-CNN的马铃薯芽眼识别方法。通过确定性重排列技术对RoIAlign操作进行改进,消除插值操作的不确定性,确保不同尺寸的候选区域在特征图上的映射的唯一性,提高了目标分类和分割的准确性和稳定性。试验结果表明,改进后的模型识别精确率为98.47%,召回率为96.99%,调和平均值F1为97.72%,平均单幅图像的识别时间为0.135s。与改进前的算法相比,识别精确率、召回率、F1值分别提升了6.46、12.01、9.36个百分点,平均单幅图像的识别时间比改进之前提升了0.004 s。改进的Mask R-CNN算法能够更好地适应马铃薯芽眼的特殊形态和不同的环境因素,有效识别出马铃薯芽眼,为马铃薯种薯智能切块装置研究奠定了良好基础。The accurate recognition of potato bud eyes stands as the prerequisite for the automatic cutting of potato seed potatoes.To improve the recognition effect of potato bud eyes and solve the long-existing problem of low efficiency of the manual cutting,a potato bud eye recognition method based on the improved Mask R-CNN was put forward.With the optimization of the RoIAlign operation by the deterministic rearrangement technology,the uncertainty of the interpolation operation was eliminated,and the uniqueness of the mapping of candidate regions of different sizes on the feature map was guaranteed,improving the accuracy and stability of target classification and segmentation.The test results indicated that the enhanced model had a recognition accuracy rate of 98.47%,a recall rate of 96.99%,a harmonic mean F1 of 97.72%,and an average recognition time of a single image of 0.135 seconds.Compared with the algorithm before improvement,the recognition accuracy,recall rate,and F1 value had increased by 6.46,12.01,and 9.36 percentage points respectively,and the average recognition time of a single image had increased by 0.004 seconds.The improved Mask R-CNN algorithm could better acclimate to the special shape of potato bud eyes and different environment factors and identify potato bud eyes,paving a solid ground for the research of potato seed potato smart cutting devices.

关 键 词：Mask R-CNN算法 马铃薯芽眼识别 自动化切块 深度学习 智能农机装备 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]