TEE Protected Drone Inspection for Critical Infrastructure: Securing Edge Analytics and Trust at Scale

Abstract: Operators of critical infrastructure are required to inspect distributed assets on a regular basis to ensure safety and reliability. These assets span a wide range of domains, including bridges and roads, rail corridors, pipelines, power transmission lines, solar farms, wind turbines, telecom towers, ports, and large industrial facilities. Traditionally, inspections have been performed by field workers. While expert technicians are indispensable, manual patrols expose personnel to electrical, physical, and environmental hazards, which may constrain inspection frequency, coverage, and data quality. In recent years, many organizations have started to utilize drone technology to enhance safety, improve efficiency, and reduce operational costs. Unmanned aerial vehicles (UAVs) are a promising productivity amplifier: a pilot can dispatch a drone to nearby functional locations, collect high-resolution RGB and thermal imagery, run analytics to identify assets and assess their conditions, then send the data back for gatekeeping, desktop analysis, and master data updates. Yet the edge-centric workflow that makes drones efficient (e.g., sensing, filtering, and triage on board) also expands the attack surface. A compromised UAV or malicious insider can exfiltrate or manipulate observations and processing results, eroding trust and undermining downstream analytics.This paper proposes a trusted execution environment (TEE)–protected surveillance solution for UAVs that enforces end-to-end confidentiality and integrity across the entire data lifecycle. We isolate acquisition and pre processing inside a hardware rooted enclave, use encryption for sealed storage and in flight telemetry, and expose a remotely attested interface to the ground system. The design has four pillars. (1) Remote attestation for trusted loading: Before a flight, the ground station verifies that only vetted processing binaries are measured and loaded into the enclave. Mission keys are released only after successful attestation, binding computation to a specific enclave identity and configuration and ensuring information integrity. (2) Confidential processing in the enclave: Sensor data collection, feature extraction, data processing and analytics execute entirely within the enclave. Encryption keeps data confidential at rest and in transit while outside of the enclave, preventing leakage even if the host OS is compromised. (3) In enclave anomaly screening: To resist data poisoning and spoofed inputs, we deploy an autoencoder based anomaly detector inside the enclave. Trained on nominal RGB/thermal signatures from prior inspections, the model reconstructs incoming observations and flags high residual frames indicative of adversarial, fake, or tampered sensor data (e.g., GAN synthesized frames, injected hot spots, or replays). Flags and reconstruction scores are attached to each packet to inform downstream trust decisions. (4) Enclave signed results for verifiable provenance: Processed outputs are signed inside the enclave with keys cryptographically bound to the attested enclave instance. The ground station verifies these signatures and associated quotes, ensuring results arrive intact and unhampered and establishing a non repudiable chain of custody from edge to cloud.We designed the framework and security protocols of this system and studied different use cases to understand the feasibility of the proposed design. Our findings show that the framework increases operational efficiency in infrastructure surveillance and simultaneously enforces lifecycle security for surveillance data across acquisition, transport, and processing.

Keywords: UAV, Trusted Execution Environment, Edge Computing, Cyber Security, Critical Infrastructure, Inspection

DOI: 10.54941/ahfe1007075

Cite this paper:

​