Cacti, belonging to the Cactaceae family, are renowned for their remarkable diversity, ranging from small globular shapes to towering columnar forms, thriving in arid environments worldwide. This morphological variability, while ecologically significant, poses a substantial challenge for accurate breed identification due to overlapping spine patterns and growth forms, making manual classification time-consuming and error-prone, especially for large-scale botanical or horticultural applications. To overcome these limitations, this study proposes an automated image-based classification system for five cactus breeds: Ball Cactus, Bunny Ear Cactus, Chin Cactus, Easter Lily Cactus, and Starfish Cactus using deep learning techniques. The research leverages a primary dataset of 2,700 images, collected through fieldwork and split into 60% training, 20% testing, and 20% validation sets, with preprocessing including resizing and augmentation to enhance model robustness. Among evaluated models—baseline CNN, CNN + SVM, CNN + Fusion, CNN + Attention, and CNN + Transfer Learning—the proposed CNN + Transfer Learning model, utilizing VGG16, achieved the highest accuracy of 96.83% and a loss of 34.85%, demonstrating superior performance across precision, recall, and F1-score metrics. This scalable approach supports digital plant identification, ecological monitoring, and smart agriculture, with future enhancements planned to include larger datasets and transformer-based models.