近年来,随着图像处理能力的提高和图像采集系统的发展,基于计算机视觉的图像分析方法在包括土壤科学在内的许多领域引起了极大的关注和推广应用.通过多种光谱组合形式的相机采集动态或静态土壤图像,然后通过相应计算机程序对其进行分类,可以进一步对土壤特性,包括水分、温度及有机碳等进行评估.为进一步加强农业种植能力,需要进一步预测农业生产中土壤的未来参数,高质量的土壤时序预测对于相关研究及农业生产具有重要意义.本研究综述了深度学习在土壤监测中的应用,指出深度学习是一种基于人工神经网络的表征学习算法,可以从各种类型的地理空间影像和数据中提取有意义的信息,在归纳总结了深度学习概念、特征的基础上,就深度学习在土壤监测中的应用及研究现状进行整理和综述,提出基于该预测方式可以给相关研究人员及农业和管理人员以较低成本的数据实现对农业生产中土壤特性的动态监测和预测,从而更好地指导农业生产.最后,从数据获取和质量、计算机视觉等方面提出了展望,以期能为深度学习在土壤监测中的应用提供参考,为精细化农业种植管理、现代化工业种植及农业物联网的发展提供支撑.
In recent years,with the increase in image processing power and the development of image acquisition systems,computer vision-based image analysis methods have attracted significant attention and spread in many fields,including soil science.The acquisition of dynamic or static soil images by cameras in the form of multiple spectral combinations,which are then classified by appropriate computer programs,allows further assessment of soil properties,including moisture,temperature and organic carbon.In order to further enhance agricultural farming capabilities,further predictions of future parameters of soils in agricultural production are required,and high-quality soil time series predictions are important for both relevant research and agricultural production.This paper reviewed the application of deep learning in soil monitoring,pointed out that deep learning is an artificial neural network-based representation learning algorithm that can extract meaningful information from various types of geospatial imagery and data.This review therefore collated and summarized the application and research status of deep learning in soil monitoring on the basis of summarizing the concepts and features of deep learning,and proposed that based on this prediction method it can give relevant researchers and agricultural and management personnel to achieve dynamic monitoring and prediction of soil properties in agricultural production with lower cost data,so as to better guide agricultural production.Finally,the outlook was presented in terms of data acquisition and quality,computer vision,etc.,which is expected to provide a reference for the application of deep learning in soil monitoring and to support the development of refined agricultural planting management,modern industrial planting and the Internet of Things in agriculture.