The Design and Implementation of Flower Image Classification based on Deep Convolutional Neural Networks

In recent years, the recognition and classification of flower species have become increasingly important in fields such as botany, agriculture, and ecological conservation. This paper improves a flower image classification model utilizing deep convolutional neural networks (ResNet) by incorporating deeper residual structures, hence enhancing the model's capacity to capture intricate feature correlations effectively. Additionally, novel data augmentation techniques, including random cropping, rotation, and color alteration, are employed to enhance the model's resilience to diverse input data. During the experiment, a public dataset containing five flower categories (approximately 3,660 images) was used, and the dataset was split into training and testing sets. The ResNet model was pre-trained, and the network architecture was fine-tuned. During training, the author recorded the loss and accuracy for each epoch to track the model's progress. The experimental results demonstrate that the proposed model can efficiently and accurately recognize flower species in the dataset. Traditional image classification methods typically achieve an accuracy of around 80%, while the optimized ResNet model achieves an accuracy of 90%.