With the continuous deepening of botanical research, a large amount of multimodal plant data, including leaf images, gene sequences, metabolite data, etc., has emerged. Single modality data is difficult to fully reveal information such as plant growth and development, physiological characteristics, and environmental adaptability. Therefore, multimodal biological data fusion has become a key means to deeply explore the potential information of plants and promote the development of plant science research. In view of the problems of inconsistent data format and complex feature association in plant multimodal data fusion, this paper constructs a complete algorithm framework. Through data preprocessing, feature fusion and model construction, it realizes the effective fusion of plant data of different modalities, and applies it to practical scenarios such as plant classification and pest and disease diagnosis. The experimental results show that the classification accuracy after multimodal fusion (92%) is significantly higher than that of single modal data (78% for leaf images, 80% for gene expression, and 75% for metabolite changes). This further verifies the effectiveness of multimodal data fusion in improving plant classification accuracy and provides strong support for plant science research and agricultural production.