Adaptive Multi-Stream Steganographic Intelligence Integration via Aggregated Filter Ensembles and Autonomous Deployment (AMSI-AFEAD)

Abstract

Steganographic systems must now confront deep-learning steganalyzers that routinely break single‑carrier schemes. This paper proposes the Adaptive Multi‑Stream Steganographic Intelligence Integration via Aggregated Filter Ensembles and Autonomous Deployment (AMSI‑AFEAD) framework. AMSI‑AFEAD conceals intelligence data across multiple parallel cover images, assigning a heterogeneous set of spatial‑frequency filters (Gabor, steerable Gaussian, and a light learned convolutional filter) to distinct streams. An autonomous deployment module monitors channel health (latency, packet loss, estimated adversarial presence) and dynamically redistributes payload fragments to keep per‑stream detectability below a predefined threshold. The system also enforces a multi‑objective policy that guarantees graceful degradation if up to half the streams are intercepted. Experiments on the BOSSBase‑1.01 and BOWS‑2 collections show that AMSI‑AFEAD raises the detection error rate by up to 14 percentage points against SRNet and Xu‑Net compared with the most secure single‑stream adaptive method, while achieving a Multi‑Stream Imperceptibility Gain of 3.5 dB and an Adaptive Security Index of 0.93. These results confirm that heterogeneous filter ensembles combined with autonomous payload routing offer a practical pathway to next‑generation covert communications.