Abstract: With development of internet of things and artificial intelligence, integration of deep learning technology with agricultural field has become increasingly mature．It has shown excellent performance in processing large-scale, high-dimensional agricultural data, and has become an important technological tool for smart agriculture development．Focused on smart agriculture, key roles of deep learning was elaborated from perspectives including pest and disease identification, crop localization, and crop yield prediction．Its shortcomings and deficiencies were analyzed, and solutions were proposed．Aiming to further promote integration of deep learning and smart agriculture, and to improve agricultural sector's development toward high-quality and high intelligence．