Plant Seedling Recognition Using Swin-Transformer Network Architecture

The study proposes a novel method for plant seedling image recognition using the Swin-Transformer network architecture, a convolutional neural network architecture developed by the Microsoft Research Institute. The aim of the study is to overcome the limitations of traditional manual methods of plant seedling classification through machine-based recognition to achieve precise identification of crops during field nursery, fertilizer management, and enhance crop yield and quality. The proposed approach leverages the Swin-Transformer network and various data processing techniques to achieve higher accuracy and generalization ability for fine-grained seedling classification. The study includes extensive experiments on relevant datasets using a high-performance GPU environment and compares its method with other state-of-the-art approaches, which demonstrate better classification accuracy results. Besides plant seedling recognition research, the paper also has practical applications. Finally, the paper evaluates the proposed method using commonly used machine learning evaluation metrics to assess model performance.