基于深度学习的番茄苗分级检测研究  

A study on the clasification of tomato seedlings based on deep learning

作　　者：静茂凯 孔德刚[1] 张秀花[1] 王鹏 袁永伟[1] 冯生 李江涛 JING Maokai;Kong Degang;ZHANG Xiuhua;WANG Peng;YUAN Yongwei;FENG sheng;LI Jiangtao(College of Mechanical and Electrical Engineering,Hebei Agricultural University,Baoding 071001,China)

机构地区：[1]河北农业大学机电工程学院,河北保定071001

出　　处：《河北农业大学学报》2023年第2期118-124,共7页Journal of Hebei Agricultural University

基　　金：河北省重点研发计划(20327207D);河北省引进留学人员资助项目(C20200336)。

摘　　要：分级是保证工厂化穴盘育苗质量的重要环节,本文根据番茄穴盘苗分级检测过程存在智能化水平低的问题,基于Darknet框架YOLOv3-Tiny卷积神经网络进行了算法改进。改进的算法进行了K-Means++聚类,增加YOLO检测层的数量,引入不同的SPP结构和CIOU损失函数。实验表明,与YOLOv3-Tiny算法相比,改进的YOLOv3-Tiny算法对番茄苗分级检测的mAP指标提高了9.8%,对壮苗的检测准确率为98.1%,无苗的检测准确率为94.80%,弱苗的检测准确率为93.62%。该算法能够对番茄苗的分级检测起到良好效果,为番茄穴盘苗高效识别提供了参考。Clasification is an important link to ensure the quality of factory plug seedlings.According to the low level of intelligence in the tomato plug seedling grading and detection process,the algorithm was improved based on the Darknet framework YOLOv3-Tiny convolutional neural network.The improved algorithm perfonned K-Means++clustering,increased the number of YOLO detection layers,and introduced different SPP structures and CIOU loss functions.Experiments showed that the improved YOLOv3-Tiny algorithm increased the mAP index of tomato seedling classification by 9.8%compared with the results obtained by YOLOv3-Tiny algorithm.The detection accuracy rates were 98.1%,94.80%and 93.62%for strong seedlings,non-seedlings and weak seedlings,respectively.The algorithm can play a good effect on the classification and detection of tomato seedlings,and provided a reference for the efficient identification of tomato seedling plug seedlings.

关 键 词：番茄穴盘苗分级 分级检测 深度学习 YOLOv3-Tiny 

分 类 号：S24[农业科学—农业电气化与自动化] TP389.1[农业科学—农业工程]