基于分层卷积神经网络的牧草种子识别模型  

FORAGE SEED RECOGNITION MODEL BASED ON HIERARCHICAL CONVOLUTIONAL NEURAL NETWORK

作　　者：王欣宇 马玉宝[2] 潘新[1] 闫伟红[2] WANG Xinyu;MA Yubao;PAN Xin;YAN Weihong(College of Computer and Information Engineering,Inner Mongolia Agricultural University,Hohhot 010011,China;Institute of Grassland Research,Chinese Academy of Agricultural Sciences,Hohhot 010010,China)

机构地区：[1]内蒙古农业大学计算机与信息工程学院,呼和浩特010011 [2]中国农业科学院草原研究所,呼和浩特010010

出　　处：《内蒙古农业大学学报（自然科学版）》2021年第5期89-93,共5页Journal of Inner Mongolia Agricultural University(Natural Science Edition)

基　　金：国家自然科学基金项目(61562067,61962048)。

摘　　要：针对目前识别牧草种子存在主要依赖于耗时费力的人工种植识别、自动化程度低等问题,本文提出并构建了基于深度学习的多层卷积神经网络禾本科牧草种子分类识别模型。通过改进单层卷积层优化提取牧草种子深度特征,并通过softmax种子分类器对10类禾本科牧草种子进行训练与分类,同时与其他分类识别方法进行比较分析。结果表明:本模型对10类纹理特征极为相似的禾本科牧草种子图像的识别率可达到91.67%,比其他方法平均高出7%~44%,并且模型具有很好的鲁棒性。验证了深度学习在牧草识别中的可行性,为牧草管理数字化提供了参考。Aiming at the problems that the current forage seeds identification mainly relies on time-consuming and labor-intensive artificial planting identification,and the low degree of automation,a multi-level convolutional neural network recognition model for the gramineous forage seeds based on the deep learning technology was constructed in this paper.The depth characteristics of the forage seeds were optimized by improving the single-layer convolution layer,and 10 types of gramineous forage grasses were trained and classified by the softmax seed classified.At the same time,it was compared with other classification methods.The results showed that the recognition rate of this model for 10 kinds of gramineous forage seed images with similar texture features could reach 91.67%,which was 7%to 44%higher than other methods,and the model had good robustness.It verified the feasibility of the deep learning applied in forage identification,and provided a reference for the digitization of pasture management.

关 键 词：图像识别 种子识别 卷积神经网络 深度学习 

分 类 号：TP319[自动化与计算机技术—计算机软件与理论]