摘要
为提高麦穗识别技术检测效率和检测精度,以GWHD数据集为基础数据集并进行数据扩充,针对现有的Faster R-CNN深度学习算法模型做了一定改进,在特征提取网络中用ResNet50网络替代了VGG16网络,并在ResNet50网络引入BiFPN加权融合单元.同时在模型中加入了K-means聚类算法,通过对目标框进行聚类从而得出9个先验框的长宽比,以此来替代原算法中固定的先验框长宽比例,以更好地适应训练.经过改进,模型的泛化能力得到增强,mAP准确率达到91.13%,相比改进前提高了0.54%,同时单张检测时间也缩短了0.1 s,提高了检测效率,验证了改进模型的可行性.
In order to improve the detection efficiency and accuracy of wheat spike recognition technology,the GWHD data set was taken as the basic data set and the data was expanded in this study.Some improvements were made to the existing Faster R-CNN deep learning algorithm model,and ResNet50 network was used to replace VGG16 network in the feature extraction network.BiFPN weighted fusion unit is introduced in ResNet50 network.At the same time,K-means clustering algorithm was added to the model,and the aspect ratio of nine prior frames was obtained by clustering the target frames to replace the fixed aspect ratio of prior frames in the original algorithm,so as to better adapt to training.After the improvement,the generalization ability of the model is enhanced,and the mAP accuracy rate reaches 91.13%,0.54%higher than before the improvement.Meanwhile,the detection time of single sheet is reduced by 0.1 seconds,which improves the detection efficiency and verifies the feasibility of the improved model.
作者
徐博文
童孟军
XU Bowen;TONG Mengjun(School of Mathematics and Computer Science,Zhejiang A&F University,Hangzhou 311300,China;Zhejiang Provincial Key Laboratory of Forestry Intelligent Monitoring and Information Technology Research,Zhejiang A&F University,Hangzhou 311300,China)
出处
《湘潭大学学报（自然科学版）》
CAS
2022年第4期48-59,共12页
Journal of Xiangtan University(Natural Science Edition)
