Contents I. Introduction
Pattern recognition has achieved considerable success in remote sensing imaging, traffic monitoring, biomedicine, and so on. In a harsh environment, such as atmospheric turbulence disturbance and weak illumination, target recognition based on traditional radar imaging or optical imaging is difficult to work due to the weakness of transmission signal or the optical diffraction effect. Quantum imaging shows better anti-interference performance than traditional methods because of the entanglement and nonlocality of light quantum. Due to the process of quantum imaging that separates the detection process from the imaging process, the target recognition technology based on non-linear optical methods has received much attention [1]–[3]. Unfortunately, these methods are easily affected by mechanical fluctuation, beam drift, and environ-mental noise, resulting in measurement errors or deviations, making it difficult to extract target feature information. Indeed, a deep learning method such as VGG can be applied to extract and classify quantum images. However, the recognition accuracy and robustness of the existed deep learning method should be reconsidered. Since it is difficult to recognize different types of images with very similar appearance features caused by the serious destruction of the local information of the target by noise, this paper proposes a target recognition method of entangled light quantum imaging based on the designed two-stream feature fusion convolutional network (TSFFCNet). The results show that TSFFCNet can recognize the target stably and efficiently even in poor quantum imaging conditions.
