Fast and blur-resilient target classification enabled by single-pixel detection

Rapid and reliable target classification is essential in resource-constrained vision applications, yet conventional imaging-based approaches suffer from severe performance degradation under optical aberrations, defocus, and motion blur. Single-pixel detection has recently emerged as a promising solution for low-cost and high-speed sensing, but existing methods mainly address motion-induced degradation and still rely on image reconstruction or computationally intensive recognition pipelines. In this Letter, we present a fast and blur-resilient classification framework that directly extracts blur-invariant features from single-pixel measurements using only seven fixed DMD modulation masks. The proposed system bypasses the traditional "image-then-recognize" paradigm and instead enables a direct "measure-to-recognize" workflow without the need for image processing or neural network training. Extensive simulations and experiments demonstrate 98.96% recognition accuracy on the test dataset at an update rate of 2.551 kHz under various degraded imaging conditions. The proposed framework provides a simple and efficient solution for blur-robust recognition and opens a new, to the best of our knowledge, avenue for high-speed optical intelligence in challenging imaging environments.