Disease identification is one of the important applications of computer vision technology in agriculture, playing a crucial role in timely detection and early prevention of plant diseases. In recent years, with the continuous evolution of disease identification methods, there has been a significant improvement in disease recognition performance. However, challenges still exist in extracting disease characteristics under natural conditions and differentiating disease severity. To explore new approaches for disease identification beyond existing methods, the current research status of disease identification and disease severity identification was analyzed. Furthermore, a comprehensive comparison and study of plant disease recognition techniques were conducted, considering visual feature types and learning methods. It pointed that deep modeling is the mainstream approach for plant disease identification. Additionally, combining multiple sources of information and utilizing different machine learning techniques are important means to improve plant disease recognition. The performance of different recognition methods was evaluated and analyzed using popular datasets. Finally, future development directions were outlined.