| 摘要: |
| 深度学习是目前机器学习领域最前沿和最具前景的技术，本研究采用归纳总结法，介绍了深度学习的特征及与传统机器学习的区别，归纳和梳理了深度学习在大田种植中的应用现状。结果表明：1）深度学习在大田种植中的应用初现端倪，主要集中在作物的识别与分类、农业遥感影像应用、土壤环境监测、农业场景识别等；2）采用的主要模型有卷积神经网络（CNN）、自编码（AE）、深度置信网络（DBN）、堆栈自编码（SAE）、全卷积神经网络（FCN）、深度神经网络（DCNN）等，其对各领域的分类与识别精度均有提高；3）目前存在的主要问题是标注数据缺乏，尤其在遥感图像分类领域，普遍采用了迁移学习、数据增强、微调等技术来解决标注数据缺乏的问题。随着大田种植领域数据的增长以及信息技术的快速发展，基于深度学习和多源异构数据的作物识别与分类、作物长势监测、病虫害预测预警、农作物产量预测、果树花朵及果体识别、水果质量及产量的优化控制等将会获得较快发展。 |
| 关键词: 深度学习 机器学习 大数据 智慧农业 大田种植 |
| DOI：10.11841/j.issn.1007-4333.2019.01.16 |
| 投稿时间：2018-04-26 |
| 基金项目:国家自然科学基金项目（61471133） |
|
| Current situation and prospect of deep learning application in field planting |
|
GUO Xiangyun1, TAI Haijiang2
|
| (1.School of Information Management, Beijing Information Science and Technology University, Beijing 100192, China;2.College of Information and Management Science, Henan Agricultural University, Zhengzhou 450046, China) |
| Abstract: |
| Deep learning (DL) is currently the most advanced and promising technique. In order to analyze the current status of DL application and put forward potential key research point in field planting, literature is collected and combed based on inductive and summary method. Firstly, the characteristics of DL and difference between DL and traditional machine learning (ML) are introduced. Secondly, articles are surveyed and summarized in order to see the current progress of deep learning in agriculture. The results show that:1) The application of DL in field planting is in the early stage and interests are focused on crop detection and classification, agricultural remote sensing image classification, soil moisture monitoring and simulation, anomalies detection in agricultural field; 2) Main models employed are Convolutional Neural Networks (CNN), Autoencoders (AE), Deep Belief Networks (DBN), Stacked Autoencoders (SAE), Fully Convolutional Networks (FCN), Deep Convolutional Neural Networks (DCNN), which improve classification and recognition accuracy; 3) The main obstacle for the application of DL in agriculture is the lack of quantities of labelled data, especially in remote sensing. With the development of information technology and large volume of collected data, DL, driven by multi-source and heterogeneous data, fast development in many fields, such as crop identification and classification, monitoring of growth condition, diseases and insect pests forecasting and warning, crop yield forecasting, detection and identification of flower and fruit, optimization and control of fruit quality and quantity will be achieved. |
| Key words: deep learning machine learning big data smart farming field planting |
