摘要
深度学习是一种新兴的图像处理和数据分析技术,其中深度卷积网络在处理图像、视频、语音和音频方面取得突破性进展,其在农业领域的应用引起广泛关注。对采用深度学习技术的39项农业病虫害检测识别研究成果进行研究,分析其数据来源、预处理和增强技术、应用领域、采用的模型和框架、性能指标,并与其它研究方法作对比。研究结果表明,深度学习具有良好的自动特征提取功能,提供了更好的分类效果,优于传统的机器学习方法,且数据采集的多样性、数据规模和完整性对深度学习性能有重要影响。
Deep learning is a new image processing and data analysis technology,Deep convolution network has made breakthroughs in image,video,voice and audio processing,and its application in agriculture has attracted wide attention.In this paper,30 research results of agricultural pest detection and identification using deep learning technology are reviewed,and their data sources,pretreat⁃ment and enhancement techniques,research application fields,models and frameworks adopted,performance indicators and compari⁃son with other research methods are analyzed and summarized.The results show that deep learning has good automatic feature extrac⁃tion function,provides better classification effect,and is superior to traditional machine learning methods.At the same time,it is found that the diversity of data acquisition,the size and integrity of data have important impacts on the performance of deep learning.
作者
边柯橙
杨海军
路永华
BIAN Ke-cheng;YANG Hai-jun;LU Yong-hua(College of Information Engineering,Lanzhou University of Finance and Economics,Lanzhou 730030,China)
出处
《软件导刊》
2021年第3期26-33,共8页
Software Guide
基金
甘肃省电子商务技术与应用重点实验室(兰州财经大学)开放基金项目(2018GSDZSW63A14)
甘肃省自然基金项目(18JR3RA216)。
关键词
深度学习
病虫害
图像识别
特征提取
智能农业
deep learning
pests and diseases
image recognition
feature extraction
intelligent agriculture
