Multi-Modal Automatic Target Recognition for Anti-Ship Missiles with Imaging Infrared Capabilities

We propose a novel Automatic Target Recognition architecture suitable for Anti-ship missiles with Imaging Infrared (IR) capabilities. Our solution combines the description capability of the suggested modified Zernike moments along with two multi-layered Artificial Neural Networks. A unique and novel advantage of our solution is its multi- modal ATR capability. Indeed although the missile acquires the target's IR signature, template matching relies on commercially available grayscale vessel silhouettes. Additional advantages include incorporating an extremely small database that has a single template per target, scale and azimuthal rotation invariance and real-time processing performance. We evaluate our architecture on simulated scenarios created in CounterSim and achieve 94% target recognition in 44ms. Focusing on the head/rear view, which are the hardest to recognize, the proposed method outperforms both Convolutional Neural Networks and state-of-the-art local feature based description techniques that are widely used in image matching and recognition.