Accurate and timely detection of crop diseases play a pivotal role in promoting sustainable agriculture and safeguarding food supplies. Traditional methods based on manual inspection are labor-intensive, time-consuming, and prone to human oversight. Although deep learning techniques have demonstrated substantial potential, they often face limitations in balancing local feature extraction and global contextual understanding, especially under hardware constraints typical of field environments. In response to these limitations, we introduce UFOViM, a hybrid architecture that merges MambaVision with the Unit Force Operated (UFO) mechanism for precise diagnosis of grape and corn leaf diseases. MambaVision enhances visual representation through sequential state-space modeling, enabling it to capture both fine-grained and holistic patterns. Meanwhile, the UFO mechanism incorporates a computationally efficient self-attention scheme with linear complexity, reducing traditional quadratic costs to O(n). We validate UFOViM on two widely adopted datasets---the Augmented Grape Disease Detection dataset and the CD&S dataset. The results indicate that UFOViM attains classification accuracies of 90.9% for grape leaves and 87.8% for corn leaves, outperforming leading models by 3.2% and 2.5%, respectively. Furthermore, the proposed model reduces computational load by 35% and memory usage by 28%, highlighting its suitability for deployment in resource-limited agricultural settings. This study demonstrates the advantage of combining sequential visual encoding with lightweight attention mechanisms for real-world plant disease detection tasks.