Spectrum classification using convolutional neural networks for a mini-camera detection system

A mini-camera is one of several emerging cat-eye devices featuring tiny lenses with diffracted retro-reflections. It is hard for traditional active laser detection systems to identify a mini-camera because of their weak reflection. This paper proposes an anti-camera system with a spectrum-based convolutional neural network algorithm to recognize the profile features of the retro-reflection images captured by the system. The network was trained with the spatial spectra of local datasets and uploaded onto the embedded device. The results of several indoor experiments demonstrate that the system reached high accuracy in real-time detection, even with various types of interference.