Fig. 1. (a) Conventional single-pixel imaging method (including array sensor scanning imaging and single-pixel imaging/classification); (b) Single-pixel high-efficiency sensing method^[37]

Fig. 2. (a) A microfluidic system for cell sorting; (b) Processing of detected electrical signals^[78]

Fig. 3. Non-imaging target recognition algorithm based on projection matrix and IMED ^[45]

Fig. 4. (a) High efficient classification system based on the convolutional network^[52]; (b) Lohit’s experiment results on MNIST and ILSVRC2012^[52]; (c) Adler’s experiment results on MNIST^[86]

Fig. 5. System setup with nonlinear optics^[56]

Fig. 6. (a) "Encode-decode" network structure; (b) Classification results^[38]

Fig. 7. (a) Optimized single-pixel sensing method with optimal binarized patterns; (b) Classification results^[88]

Fig. 8. (a) Sampling rate-adaptive sensing network’s structure; (b) Classification results^[67]

Fig. 9. Description of non-reconstruction related feature estimation for advanced visual reasoning problems (taking face recognition as an example)^[42]

Fig. 10. Proof-of-concept setup for single-pixel multi-character recog-nition and the result of license plate recognition at a 5% sampling rate^[37]. (a) The proof-of-concept setup; (b) The ex-perimental results of various license plates

Fig. 11. (a) Imaging-free single-pixel scene segmentation system framework; (b) Comparison of the experimental results of conventional imaging first and then perception methods at 1% sampling rate^[70]

Fig. 12. Target tracking results in the x-O-z plane (a), y-O-z plane (b)and the corresponding three-dimensional targets(c)^[62]