Plant Species Recognition and Classification Algorithm Based on Deep Learning

In the fields of biodiversity conservation, agricultural management, ecological monitoring, etc., the value of plant species identification and classification is immeasurable, but traditional methods are limited by many factors such as a wide variety of species, high morphological similarity, and environmental interference, and the recognition and classification performance is insufficient. This paper introduces Vision Transformer (ViT) and combines it with a transfer learning strategy to solve this problem. The paper first collects plant image data and performs data preprocessing, and then uses these image data to pre-train the ViT model, applies the pre-trained model to the plant species identification task, and fine-tunes it through a transfer learning strategy. In the fine-tuning process, this paper uses plant image datasets to train the model. At the same time, data enhancement techniques such as random cropping, rotation, and flipping are used to increase the model's capabilities. After the training is completed, the paper conducts a comprehensive evaluation of the model and further optimizes the model based on the evaluation results. The classification accuracy of the ViT model on multiple plant species image sets is generally high, especially in some image sets, reaching 99.9% accuracy. The classification time of the ViT model in all test image sets is lower than that of YOLOv5, showing faster processing speed. The mAP value of the ViT model is also higher than that of the YOLOv5 model, indicating that ViT has higher detection accuracy and recognition ability in plant species image classification tasks. Therefore, this paper provides a new and efficient solution for ecological research and biodiversity conservation.