High Speed Single Pixel Imaging Measurement Value Classification Through Multi-Layer Perceptron

The slow reconstruction process of single-pixel imaging makes it difficult to apply to dynamic scenes, and it also causes serious computational waste for the reconstruction of objectless data. To address this, we propose a method of using multi-layer perceptron to achieve high-speed screening and object detection of single pixel imaging data. This method achieves a detection rate of 2500 frames per second (fps) and an accuracy of 94.98% on the MNIST dataset. The classification accuracy of our proposed method is 87.8% at 0.77% sampling rate and in real scene tests, and it maintains good performance under 10% random noise.