Inductive Conformal Prediction Enhanced LSTM-SNN Network: Applications to Birds and UAVs Recognition

Deep learning stands out as a potent state-of-the-art technique for target recognition. Unfortunately, the trustworthiness and reliability of deep learning networks encounter challenges in radar target recognition. In this letter, an inductive conformal prediction (ICP) enhanced long short-term memory spiking neural network (LSTM-SNN) is proposed. It integrates with the concept of conformal prediction in statistical learning theory and deep learning, and is applied to birds and drones' recognition with radar. The proposed method can provide good recognition results for drones and birds with supplying confidence and credibility for each identification, and it yields a confidence interval containing the true value of the estimated at the desired confidence level, such as 98%. The benefits of the ICP enhanced LSTM-SNN method were demonstrated with the bird detection datasets obtained by radar in the airport.