Detection of plant health in the early stages with precise methods is critical to precision agriculture, as it allows early preventive action and prevents the expansion of the disease to large areas. The research introduces a deep learning-based system that performs automatic classification into 3 classes: healthy, unhealthy, and dead. The dataset is based on high-resolution images of the Money Plant, Spider Plant, Snake Plant, and Bamboo Plant under distinct illuminations and with different backgrounds to facilitate accurate and generalizable identification. The research utilizes YOLOv8, YOLOv9, and YOLOv11 to determine their effectiveness in plant health assessment through an extensive performance analysis. The experimental results demonstrate that YOLOv11 achieves better classification performance than its previous versions and delivers improved precision detection robustness and recall. The evaluation includes TensorFlow Lite analysis to address the issue of running real-time inference on mobile devices with limited resources. TensorFlow Lite maintains slightly less accurate detection results than YOLO models but still provides exceptional speed and computational performance which suits its application for mobile-based diagnostic tools and greenhouse surveillance. The real-time inference functions of the proposed models have turned them into adaptive components suitable for agricultural applications in intelligent ecosystems, including greenhouse monitoring and mobile diagnostics for farmers. Deep learning technology shows great potential to automate plant disease detection according to this study which leads to reduced manual processes alongside increased agricultural productivity benefits alongside environmental sustainability in farming. Further development will involve enlarging the available data collection as well as adding more plant species into the system and deploying it through agricultural monitoring systems on a large scale.