Computer Vision as a Measurement Channel in an AIoT Cyber-Physical Pond Aquaculture System: A System Analysis Approach

Abstract

Pond aquaculture is constrained by labor costs, environmental variability, and the need for stable water quality under uncertainty. IoT sensor networks provide continuous measurements of key water parameters, but they remain “blind” to fish behavior and stock conditions that affect feeding efficiency and early warning of adverse events. This paper extends a cyber-physical pond aquaculture system by introducing computer vision (CV) as a dedicated measurement channel within a closed-loop AIoT (AI + IoT) architecture. From a systems analysis perspective, we formalize the system boundaries, functional modules, data flows, and quality-of-service/quality-of-data constraints for real-time operation. A unified observation model is proposed that synchronizes sensor telemetry and video-derived metrics (fish count, activity level, spatial distribution) via timestamp alignment and confidence-aware data fusion. The fused state estimate drives a hierarchical control logic integrating adaptive regulation of aeration/feeding with safety rules and fallback modes under degraded sensing. A scenario-based evaluation framework is described, including environmental disturbances, sensor faults, video occlusion, and network delays. Illustrative results demonstrate that the vision-enhanced system reduces time spent in unsafe low-oxygen conditions and improves overall water quality indices compared to a sensor-only baseline, while meeting real-time constraints. The approach provides a high-level blueprint for designing and validating AIoT-enhanced aquaculture systems using computer vision as a new measurement modality.