Crop losses caused by disease, water stress, and delayed field intervention remain a major challenge for small and medium farmers. Conventional advisory systems often depend on manual inspection or cloud-only diagnosis, which can be slow in rural environments where connectivity is limited. This paper proposes a multimodal edge-intelligence framework that combines leaf-image analysis, soil-moisture sensing, weather context, and lightweight decision rules to provide early crop disease detection and irrigation advisory. The system uses a compact convolutional neural network for visual symptoms and a sensor-fusion module for environmental risk estimation. By running inference near the field, the framework reduces latency and protects farm data while still supporting periodic cloud synchronization. Simulated evaluation shows 91.8% disease classification accuracy, 16.4% water saving, and faster advisory delivery compared with image-only and rule-based baselines.