Machine Learning Isolation Forest-Based Target Detection Algorithm for Airborne Radar

While the advantages of airborne radar are widely recognized, it is prone to target overwhelm by clutter during detection tasks, and most traditional detectors rely on clutter mathematical models to obtain detection thresholds. This paper proposes a new machine learning isolation forest-based airborne radar target detection algorithm, fully exploiting the advantages of ensemble decision trees. The proposed algorithm first designs new feature data with feature extraction and space-time weight vector, and the new data enable separate utilization of multidimensional features, serving as inputs for the target and clutter classifier to be constructed. Then, the proposed algorithm designs a unique isolation forest-based target and clutter detector, constructing isolation forest anomaly scores for each designed feature data, and determining the presence of targets for airborne radar based on score discrimination criteria. Compared with various classical detectors, the proposed algorithm significantly enhances target detection performance, providing higher detection probabilities, and offering a new idea and direction for radar target detection. The proposed algorithm is confirmed by simulations to be effective and advantageous.