Edema of the lower extremities is a common symptom of chronic diseases such as heart failure, liver disease and renal dysfunction, and its progression helps in diagnosis and monitoring. Traditional methods of assessing edema based on visual inspection and palpation are subjective and inconsistent. The study proposes a deep learning system that uses YOLO models to detect areas of edema, image enhancement techniques to better display features, and classification models to determine severity. Random rotation, background removal and cropping of non-target regions were applied to address data imbalance. The system achieved an average classification accuracy of 87 to 93% for different degrees of edema severity, 90 to 94% for recall rate, and 93 to 97% for overall accuracy. The results confirm the effectiveness of automatic detection and classification of lower limb edema. The system has the potential to support clinical decision making and home patient care.