Flower segmentation and classification is a significant task in computer vision. It holds immense potential across multiple domains such as botany, agriculture, and horticulture. This paper addresses the problem of accurate and efficient flower recognition, highlighting its relevance and importance. A novel two-stage deep learning model U-Net + ResNeXt is proposed that integrates U-Net for flower segmentation and ResNeXt for classification. The proposed model aims to enhance fine-grained classification accuracy while effectively handling complex and cluttered floral images. In the experimentation, the U-Net + ResNeXt models performance is evaluated on Oxford 102 Flowers dataset, comparing it against single-stage ene-to-end classification models ResNeXt, VGG16, DenseNet and Custom CNN. The results demonstrate the superior capabilities of two-stage U-Net + ResNeXt approach, showcasing the importance of combining segmentation and classification for accurate flower recognition. The proposed model achieves considerable accuracy, precision, recall, and F1-Score, addressing the challenges of fine-grained classification and imbalanced datasets. The findings presented in this paper underscore the significance of advancing flower recognition techniques, contributing to applications in agriculture, botany, and environmental conservation.