基于深度自动编码器的小麦种子聚类识别方法  

Clustering recognition method of wheat seeds based on deep auto-encoder

作　　者：刘赛雄 耿霞[1] 陆虎[1] LIU Saixiong;GENG Xia;LU Hu(School of Computer Science and Communication Engineering, Jiangsu University, Zhenjiang, Jiangsu 212013, China)

机构地区：[1]江苏大学计算机科学与通信工程学院,江苏镇江212013

出　　处：《江苏大学学报（自然科学版）》2020年第3期294-300,共7页Journal of Jiangsu University：Natural Science Edition

基　　金：江苏大学高级人才基金资助项目(14JDG040)。

摘　　要：为了实现利用人工智能或机器学习方法提高农业生产效率,针对农业数据分析中特征提取和类别识别的问题,构建了一个基于深度神经网络的自动编码器,该网络不仅能够分析农业数据的固有属性特征,还能自动学习潜在的高级特征.通过在自动编码器中引入了一个高斯核聚类模块,提出了一个新的损失函数反向调节整个网络训练,使其逐步实现聚类的结果,最终实现了一种新的基于自动编码器的高斯核模糊聚类方法(AE-KFC).该聚类方法是一种基于自动编码器的端到端深度神经网络学习方法.最后在农业小麦种子的数据集上进行了试验测试,相比其他的一些聚类算法,提出的聚类算法取得了较好的性能结果.结果表明:新型的机器学习算法有助于提高农业数据分析的效果,具有广泛的应用价值.To improve the efficiency of agricultural production by artificial intelligence or machine learning,an auto-encoder was constructed based on deep neural network to solve the problems of feature extraction and category identification in agricultural data analysis.The proposed network could not only analyze the features of inherent properties in agricultural data,but also could learn the potential advanced features.Based on the Gaussian kernel fuzzy clustering algorithm(AE-KFC),the new auto-encoder was realized by adding a Gaussian kernel clustering module into the auto-encoder,and a new loss function was used to inversely adjust the whole network training to obtain the clustering results gradually.The proposed clustering method was an end-to-end deep neural network learning method based on auto-encoder.A great number of experiments were conducted on the dataset of agricultural wheat seeds.The results show that compared with other clustering algorithms,the proposed clustering algorithm has better clustering performance.The new machine learning algorithm can improve the effect of agricultural data analysis and has extensive application value.

关 键 词：小麦 自动编码器 聚类 农业数据分析 机器学习 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]