Plant Identification and Classification using CNN

The difficulties in obtaining high classification accuracy while preserving computational efficiency are brought to light by the expanding need for automated plant recognition systems. In order to solve the challenge of classifying leaves into the five categories of tree, herb, shrub, climber, or creeper, this paper evaluates three different deep learning methods. First, we use Vision Transformer (ViT) to extract complex image patterns; second, we use ResNet50 to perform reliable feature extraction; and third, we combine MobileNetV2 and EfficientNetB0 to achieve even greater efficiency. Using cosine simi- larity to find the best match, the methodology extracts features from a custom dataset of labeled leaf images. Main results show that ResNet50 and ViT also give good classification performance, and that the MobileNetV2-EfficientNetB0 approach offers a competitive balance between efficiency and accuracy,. While ViT and ResNet50 are also effective, the main finding is that combining MobileNetV2 and Effi- cientNetB0 produces the most balanced results EË™ xtending the dataset and optimizing the models will be the main goals of future research to enhance the classification of species.