EfficientNet-B7 is an efficient deep convolutional neural network architecture, belonging to the EfficientNet series of models. This model, through systematic research on model scaling methods, proposes a compound scaling technique, and simultaneously optimizes the network's depth, width, and input resolution, thereby achieving a better balance between accuracy and computational efficiency. This paper elaborates on the working principle of EfficientNet-B7 in detail. A flower dataset with 104 categories was downloaded from Google Cloud Server, and EfficientNet-B7 was introduced to implement flower classification. To reduce model overfitting, the DropPath regularization term was added after the loss function. Through validation and testing, EfficientNet-B7 can effectively classify all flowers successfully, with a success rate reaching 100%. The addition of the DropPath regularization term can effectively reduce training time and improve network communication efficiency. Experiments show that the flower classification research based on EfficientNet-B7 is practical and effective, and it is of great significance to the study of flower classification and recognition.