DeepEdge-AD: Deep Anomaly Detection in Edge Computing-Driven Leaf Extraction Machines

This study introduces DeepEdge-AD, a novel framework for real-time anomaly detection in industrial leaf extraction processes using IoT and edge computing.It employs a forecasting generative adversarial network for anomaly detection (ForGAN-AD), a deep learning model that integrates probabilistic forecasting with GANs to detect anomalies in streaming sensor data.Key innovations include (1) a hybrid architecture combining statistical methods and deep learning, (2) an edge computing implementation that achieves ultra-low latency (0.0019 seconds training time), and (3) a closed-loop feedback system for automatic parameter optimization.DeepEdge-AD outperforms traditional methods, achieving an Area Under the Curve (AUC) of 0.9720 and a macro F1 score of 0.7148 on the Moringa leaf dataset.Validation on public datasets (NYC Taxi: AUC 0.7462; AnoShift Kyoto: AUC 0.7613) confirms its generalizability, establishing it as a state-of-the-art solution for real-time anomaly detection in edge environments.