Task-oriented semantic variable compression communication scheme with salient region awareness over hollow-core fiber

This paper proposes a task-oriented semantic variable compression communication scheme based on histogram contrast (HC) and compressive sensing (CS). Saliency detection (SD) is used to dynamically assess the importance of image blocks, enabling adaptive CS sampling with variable compression ratios accordingly. The scheme achieves 8 Gb/s and 16 Gb/s orthogonal frequency division multiplexing (OFDM) optical signal transmission over 1.5 km hollow-core fiber (HCF). Experimental results show that the bit error rate (BER) meets the 7% HD-FEC (hard-decision forward error correction) threshold @3.8 × 10^-3, and both the demodulated constellation diagrams and reconstructed images maintain high clarity. All test samples yield peak signal-to-noise ratios (PSNR) above 30 dB and multi-scale structural similarity (MSSIM) values exceeding 0.9. Compared with conventional fixed-ratio block compression schemes, the proposed method achieves up to 0.5 performance gain under the semantic-MSSIM (S-MSSIM) metric. These results highlight its strong potential for future high-capacity semantic optical fiber communication systems.