Abstract: The use of deep learning to accurately identify weeds to achieve the goals of pesticide reduction, efficiency and safety has been widely concerned by researchers. Therefore, the research progress of using deep learning algorithms to identify and detect weeds of different crops in recent years is reviewed at home and abroad. The review status is summarized from the aspects of data acquisition, algorithm selection, optimization and recognition effect. It points out that the problems of timeconsuming and laborious data set establishment, low accuracy of field recognition, algorithm equipment deployment difficulty, realtime performance and accuracy cannot be both, and also proposes solutions. Meanwhile, it prospects the development trend of easy and accurate data set establishment, lightweight and robust algorithm model structure, and easy deployment of mobile devices, and methods for future applications., which  will provide a reference for the subsequent research on precise weed removal.

Key words:  precision, deep learning, weed recognition, data collection, model optimization

摘要：  使用深度学习精准识别杂草实现使用农药减量、增效、安全的目标受到研究人员的广泛关注。因此综述近年来国内外使用深度学习算法对不同作物的杂草进行识别检测的研究进展，重点从数据获取、算法选择、优化部分、识别效果等方面总结研究现状，指出其在数据集建立费时费力、田间识别精度低、算法设备部署困难、实时性与准确率不能兼得等方面的问题，并提出解决方案，同时展望数据集建立简易和准确，算法模型结构轻量和鲁棒性强，便于部署移动设备的发展趋势以及未来应用的方法，为后续杂草精准化清除研究提供参考。

关键词: 精准化, 深度学习, 杂草识别, 数据获取, 模型优化