Representation learning is a method of representing the intrinsic information of research object as a dense lowdimensional realvalued vector. The main purpose is to find a better representation of the original data. Representation learning, with its ability to extract features automatically, shows high efficiency when dealing with a large amount of artificially limited prior data. Supervised and unsupervised representation learning models have been used in the analysis of plant phenotypic data such as text, images, and 3D point clouds. With the rapid growth of data in recent years and the rapid development of genomics research, plant phenotypic research data has features like high throughput and high accuracy. Representation learning models have gained attention in the analysis of massive highdimensional plant phenotypic data. The related concepts of representation learning were briefly introduced, supervised and unsupervised representation learning models were compared and analyzed, plant phenotypic data concepts and processing methods were briefly introduced, which was mainly focused on plant species identification, pest detection and analysis, yield prediction, gene research and morphological structure phenotypic data calculation, etc.. The significance of the research application of representation learning in plant phenotypes and its problems were also discussed. Finally, the application trends of representation learning in plant phenotypes were prospected: developing representation learning models that can be applied to the analysis of different plant phenotype data; improving the realtime and accuracy of representation learning models to enhance their practicality; designing multimodal phenotypic data representation learning models that provided consistent data views for phenotypic data analysis.