This study evaluates the impact of use image classification infrastructures over a real-world application, such as grapevine diseases. For the comparison, the robustness of YOLOv11 and Faster R-CNN in detecting grapevine diseases under visual degradation, aiming to simulate the real-world image capture situation, is presented. Using the well-known PlantVillage dataset, it was shown that while both models achieve near-perfect accuracy under ideal conditions, their performance degrades when exposed to noise and blur filters. YOLOv11 proved superior, being up to 15 times faster and more resilient than Faster R-CNN. The results demonstrate the importance of training AI models with real-world datasets and generalisation in the face of environmental factors in a realistic context and explore their capabilities of classifying vineyard diseases.