A New Detector Based on Alpha Integration Decision Fusion

This paper presents a new detector method based on alpha integration decision fusion. The detector incorporates a regularization element in the cost function. This element is considered a measure of the smoothness of the signal in graph signal processing. We theorize that minimizing this term will reduce the dispersion of the statistics of the fusion, and thus improving the separation between the two hypotheses of the detection. To highlight the performance of alpha integration methods and regularization classification, two experiments are presented. The first one consists of simulated data, and the proposed method is compared with alpha integration without regularization. The second one consists of detection of ultrasound pulses buried into highly background noisy. In this latter experiment, three single classifiers were implemented: support vector machine; quadratic linear discriminant; and random forest. The results obtained from those classifiers were fused by using the mean; standard alpha integration and alpha integration with regularization. In all experiments, the advantages of the proposed method were demonstrated.