基于深度学习的农作物病害图像识别技术进展

农作物病害的无损检测和早期识别是精准农业和生态农业发展的关键。随着图像采集和图像处理技术的进步,高光谱成像等先进成像探测技术和基于深度学习的图像分析技术越来越多地应用于农作物病虫害的无损检测中。本文首先简单介绍了以深度学习为代表的图像识别技术的基本原理,然后系统地阐述了基于深度学习的先进成像技术和先进图像识别分析技术在农作物病害检测识别中的国内外研究现状,分析了其在农作物病害检测识别上存在的优缺点,如具有快速、准确率高等优点以及数据量过大处理不便等缺点,并进一步指出,利用高光谱成像和热红外成像与深度学习相结合,将成为今后研究农作物病虫害早期检测的主要发展方向。

Advances in new nondestructive detection and identification techniques of crop diseases based on deep learning

The non-destructive testing and early identification of crop diseases is the key to the development of precision agriculture and ecological agriculture. With the progress of image acquisition and image processing technologies, advanced imaging detection technologies such as hyperspectral imaging and image analysis technologies based on deep learning were increasingly used in non-destructive testing of crop pests and diseases. This article first briefly introduced the basic principles of the new non-destructive testing technology represented by near-infrared thermal imaging technology and hyperspectral imaging technology and the image recognition technology represented by deep learning, and then systematically elaborated new imaging technologies and advanced image recognition and analysis technologies. The domestic and foreign research status in crop disease detection and identification was demonstrated, and its advantages and disadvantages in disease detection and identification were analyzed, with the advantages of rapidity and high accuracy, but the disadvantage of too large data volume to handle. The research trends and development directions of non-destructive testing of crop diseases were further pointed out, indicating that the combination of hyperspectral imaging with thermal infrared imaging and deep learning will be the development direction for the early detection of crop pests and diseases.