Hydrocyclones are pivotal in mineral processing for particle classification, yet operational inefficiencies like roping can lead to significant financial losses and equipment downtime. This study explores edge computing and computer vision to monitor hydro cyclone performance in real time, addressing the challenges of remote operation with limited connectivity. Building on previous work, we evaluate Convolutional Neural Networks (CNNs) and Vision Transformers (ViTs) for classifying hydrocyclone underflow images. We tested ResNet-18 and MobileViT-V2 models for their ability to detect and classify operational statuses using a conceptual prototype to simulate hydrocyclone conditions. Results indicate that MobileViT-V2, with its advanced integration of CNN and ViT features, achieved superior performance with 100% accuracy on a test dataset. Conversely, ResNet-18 models showed strong results but experienced some performance degradation on edge devices due to quantization effects. Despite these challenges, both models demonstrated feasibility for real-time monitoring, achieving F1 scores over 92.7% for critical detection tasks. This research highlights the potential of edge-based computer vision systems for enhancing operational efficiency in mineral processing. It underscores the need for further refinement in model deployment and environmental adaptability. Future work should focus on improving edge device performance and exploring the application of ViTs in such settings.