Unmanned Aerial Systems (UAS), commonly known as drones, have seen extensive growth in industries such as surveillance, agriculture, logistics, and defense. As autonomous systems, UAS often operate in critical applications that require secure data transmission for their operation. UAS face significant cybersecurity risks, especially during data transmission between the drone, ground stations, and cloud infrastructure. Traditional encryption schemes, though secure, can be inefficient for real-time data streams in the context of limited bandwidth, computing power, and network fluctuations. Adaptive encryption can adjust the level of security depending on varying parameters such as data sensitivity, transmission conditions, and available resources, thereby balancing security with operational efficiency. Recent advancements in adaptive encryption techniques offer a promising approach to fortifying these critical data flows. Unlike traditional encryption methods that operate under static, predefined protocols, adaptive encryption is designed to be dynamic, adjusting to the variable conditions and evolving threats encountered during flight [1]. These adaptive systems react in real-time, ensuring that data transmitted during autonomous flight is consistently protected despite fluctuating environmental factors [2].
In general terms, securing UAS data is not without its challenges. The dynamic nature of autonomous flight introduces complexities such as mobility, changing network topologies, and exposure to a range of potential adversaries. These factors complicate the use of static encryption protocols, which struggle to keep pace with the rapidly changing threat landscape during flight operations. Furthermore, many UAS are resource-constrained, with limited processing power and energy capacity [3]. As such, encryption solutions must not only provide robust security but also avoid overburdening the system's operational capabilities.
This paper explores the multifaceted complexities of implementing adaptive encryption techniques for UAS, highlighting the need for innovative solutions that can adjust to the realtime demands of autonomous flight while ensuring strong data security. One critical aspect of this challenge is the latency associated with encryption processes. Delays in data encryption and transmission can severely impact operational efficiency, especially in mission-critical applications where timely decision-making is essential [4]. Therefore, addressing both the security and performance requirements of UAS is crucial to their successful deployment in real-time applications.
The increasing volume and complexity of data transmitted during UAS operations further exacerbate these concerns. As UAS technology advances, data integrity and confidentiality have become more critical, driving a need for more efficient and transparent encryption protocols [5]. Traditional encryption methods often fall short when balancing robust security with the low-latency demands of real-time operations. In response, lightweight encryption techniques are gaining attention as potential solutions, designed to minimize computational overhead and reduce encryption and decryption times, making them well-suited for time-sensitive UAS applications [6]. Fig. 1 describes UAS data security challenges.
Beyond latency concerns, adaptive encryption techniques must be able to accommodate the diverse and unpredictable environments in which UAS operate. Factors such as changes in altitude, speed, and geographic location may necessitate rapid adjustments in encryption parameters to ensure contin-Fig. 1. Graphical Representation of UAS Data Security Challenges ued protection against evolving threats. This adaptability is particularly important as UAS encounter both deliberate cyberattacks and inadvertent vulnerabilities in their communication channels, underscoring the need for sophisticated encryption schemes capable of mitigating a wide array of threats.
The integration of adaptive encryption within UAS requires striking a delicate balance between security and operational efficiency as shown in Table 1. While stronger encryption algorithms may offer improved protection, they can introduce delays that undermine the real-time performance critical to UAS functionality. Achieving this balance necessitates continuous research and development to identify strategies that optimize encryption without compromising the system's responsiveness.
TABLE I INTEGRATION OF ADAPTIVE ENCRYPTION IN UAS SECURITY Aspect Description Category Main Focus Details Technical Challenges Adaptive encryption Responds to evolving threats dynamically. Policy & Regulatory Compliance Ensuring adherence to UAS cybersecurity regulations. Operational Practices Seamless integration Works within autonomous flight operations. Compliance Requirements Security standards Aligns with national and international frameworks. Security Effectiveness Encryption balance Ensures strong encryption while maintaining realtime efficiency. a Ensuring secure integration within evolving UAS environments.
Emerging technologies, such as machine learning and artificial intelligence, offer exciting prospects for the development of adaptive encryption systems capable of learning from realtime operational conditions and dynamically adjusting encryption protocols. These intelligent systems hold the potential to enhance the resilience of UAS communication networks against an evolving spectrum of security threats [7]. However, their deployment also raises challenges related to algorithmic transparency and accountability, particularly when applied in sensitive applications.
As UAS technology continues to evolve, the demand for advanced adaptive encryption techniques grows more urgent. The development of such techniques is not solely a technical challenge; it also requires careful consideration of policy, regulatory frameworks, and operational practices that govern UAS use. The integration of adaptive encryption must align with these evolving regulations to ensure both compliance and effective security in autonomous flight operations.
The increasing reliance on UAS in diverse fields from surveillance and agriculture to telecommunications and emergency services means that the potential consequences of data breaches are more significant than ever. Adaptive encryption has emerged as a key solution to addressing these vulnerabilities, enabling encryption protocols that can dynamically adjust to changing threat landscapes. However, as these systems grow more complex, it is essential to continually refine them to balance security, performance [7], and real-time operational requirements.
Moreover, emerging technologies like blockchain, which promises to enhance the security and accountability of UAS data transmission, add both opportunities and challenges to the landscape of adaptive encryption. While blockchain's decentralized nature could offer enhanced transparency and protection, it also introduces additional complexity into the encryption strategies that must be addressed [8]. The integration of blockchain with UAS encryption requires solutions that can reconcile the strengths of these technologies while addressing the rigorous security needs of real-time data transmission. Fig. 2 depicts a simple mechanism for an adaptive encryption. In sum, the evolution of UAS technology, combined with the growing importance of real-time data security, highlights the critical need for adaptive encryption techniques that can address the dynamic and unpredictable nature of autonomous flight. Encryption techniques such as symmetric encryption (AES) and asymmetric encryption (RSA/ECC), hybrid encryption and key management can be explored for adaptive framework. By examining the interaction between encryption, emerging technologies, and regulatory considerations, this paper aims to shed light on the strategies and innovations needed to secure UAS operations in an increasingly connected and complex world. This paper will further explore how adaptive encryption can be applied to real-time UAS data transmission.
The emergence of Unmanned Aerial Systems (UAS) has revolutionized various sectors, introducing advanced capabil-ities in surveillance, reconnaissance, and autonomous flight. However, these advancements also pose significant cybersecurity risks, particularly concerning the integrity and confidentiality of real-time data flows.
Adaptive encryption has been increasingly recognized as a vital mechanism to enhance the security of UAS communication channels. Unlike static encryption models, adaptive encryption mechanisms dynamically adjust to the characteristics of the transmitted data and evolving security threats. This ensures that autonomous flight systems maintain data confidentiality even in highly volatile environments.
According to Ref. [9], randomized adaptive encryption techniques shown in Fig. 3 have demonstrated potential in securing data flows transmitted by UAS. These mechanisms not only offer a flexible approach to encryption but also respond dynamically to emerging threats, making them highly suitable for autonomous and real-time communication systems.
UAS frequently operate within Internet of Things (IoT) environments, where they interact with other connected systems, sensors, and control units. The integration of UAS into IoTbased frameworks introduces new security vulnerabilities, necessitating the implementation of robust defense mechanisms.
Reference [10] emphasize that without proper encryption models, UAS data transmissions remain vulnerable to unauthorized access, interception, and cyberattacks. Given that these systems transmit large volumes of sensitive data, ensuring their security is paramount. Similarly, Ref. [11] highlights that the real-time nature of UAS operations exacerbates security risks. Adaptive encryption, when integrated with IoT security models, offers an effective solution by dynamically responding to network changes and user requirements.
Additionally, Ref. [12] address the growing complexity of UAS-generated data traffic, which includes diverse types of information such as Navigation and flight path data, Surveillance feeds, Real-time command and control instructions. Without efficient encryption mechanisms, the risk of data breaches shown in Fig. 4 increases, particularly in military applications, disaster response missions, and industrial UAV deployments.
As UAS technology continues to evolve, researchers are developing new encryption models specifically designed to secure UAV data flows. Recent studies have explored federated data collection frameworks and blockchain-based security models to enhance privacy protection in UAS networks.
Reference [13] proposes a federated data collection system adapted for UAVs operating in maritime environments, ensuring privacy conservation and preventing unauthorized access to UAV-generated data. Likewise, Ref. [6] introduced the DFly system, which integrates blockchain technology with adaptive encryption techniques. Their model allows for public auditability while maintaining strong encryption for secure data exchange.
Another significant challenge is securing Flying Ad Hoc Networks (FANETs), which are highly dynamic and decentralized. Reference [5] provides a comprehensive review of FANET security challenges, emphasizing the importance of adaptive encryption in mitigating risks related to data interception and network resilience.
Although these recent work such as DFly [6] and the federated data frameworks [13] provide robust mechanisms for specific security concerns in UAS networks, they often focus on blockchain integration or preservation of data privacy without addressing dynamic real-time decision making. Unlike these studies, our work contributes to a dual-layered decision system that merges predefined rules and predictive learning to adaptively respond to evolving conditions in flight. Moreover, none of the existing frameworks offers integrated rule-ML hybrid selection with evaluation across operational metrics like latency, encryption type matching, and CPU bandwidth balancing. Our approach is new in its practical hybridization and real-time encryption logic automation for UAS data streams.
The proposed adaptive encryption system to secure the UAS data stream that we designed provides real-time security while maintaining efficient communication. The architecture consists of three core components: Data Flow Control, Adaptive Encryption Mechanism (Fig. 5), and Secure Transmission Layer.
Our data flow begins at the UAS sensors and control systems, which generate telemetry, video feeds, and missioncritical data. Our data is classified based on sensitivity and transmission priority. A preprocessing module analyzes network conditions and selects an appropriate encryption level.
Unlike traditional static encryption, our adaptive encryption module dynamically adjusts encryption strength based on:
• Data Sensitivity: Critical flight commands use highsecurity encryption (AES-256, RSA), while less sensitive telemetry data may use lightweight encryption (ChaCha20, AES-128).
• Network Bandwidth: If the system detects low bandwidth or high packet loss, it applies compression-based encryption to reduce overhead.
• Computational Load: The module optimizes encryption intensity based on available processing power, ensuring low-latency encryption for real-time operations.
The encrypted data is securely transmitted to ground control stations or cloud services using:
• End-to-End Encryption (E2EE): Prevents interception during transmission.
• Ephemeral Key Exchange: Uses dynamic session-based encryption keys for secure communication.
The proposed rule-based adaptive encryption algorithm intelligently selects encryption strategies based on data importance, system resources, and environmental factors. This ensures real-time security, low processing overhead, and resilience against cyber threats in autonomous UAS networks. The diagram in Fig. 6 follows a flowchart notation based on conditional branching. In the 'Data Classification' phase, the transition to either "Bandwidth & CPU Load Check" or direct evaluation of 'Mission critical data' depends on the data labeling. Data marked as high-priority or safetycritical is routed immediately for maximum security encryption. Otherwise, less-sensitive data undergoes resource check analysis before encryption. This decision path enables efficient load management while ensuring data criticality is always prioritized.
The system integrates symmetric encryption (AES, ChaCha20) for speed and asymmetric encryption (RSA, ECC) for secure key exchange. The approach is selected based on:
• Data Classification: Sensitive control data requires highsecurity encryption, while non-critical telemetry data utilizes lightweight encryption.
• Network Conditions: Adaptive switching between encryption levels optimizes performance under fluctuating bandwidth.
• Latency Requirements: Time-sensitive data undergoes low-latency encryption to meet operational constraints.
The encryption model incorporates a dynamic key management system to ensure fresh encryption for each communication cycle and trigger key rotation in case of suspected cyber threats.
1) Approach Implementation: Our rule-based adaptive encryption algorithm was run under different conditions and generated a table showing the selected encryption method and its execution time in Fig. 7.
The key Observations of our approach shows the selection decisions below.
• AES-256 + RSA was selected for high-security cases with low CPU load and sufficient bandwidth, taking slightly longer.
• ChaCha20 was used for low-latency needs, ideal when bandwidth is limited or CPU load is high.
• AES-128 was chosen as a balanced option, offering moderate security with fast performance as shown in Table II. While encryption adaptability ensures flexible defense, it is also important to contextualize its role in countering specific threats. For instance, Man-in-the-Middle (MITM) attacks are mitigated via ephemeral key exchanges, reducing key reuse and minimizing interception risks. Spoofing attacks are addressed through encrypted identity verification protocols tied to mission-critical data, ensuring authentication before control command execution. Future work will implement simulated adversarial testing to validate robustness against these threats.
TABLE II RULE-BASED ENCRYPTION METHOD PERFORMANCE Method Encryption Time (s) AES-256 + RSA (High Security) 9 × 10 -5 ChaCha20 (Low Latency) 2 × 10 -5 AES-128 (Balanced) 2 × 10 -5
Even though effective, the rule-based system lacks adaptability to dynamic network conditions and emerging security threats. To overcome this limitation, a hybrid adaptive encryption mechanism is introduced. The system initially applies rule-based encryption, but integrates machine learning (ML) to refine encryption selection over time.
Furthermore, the hybrid encryption model proposed in this work is designed to address the limitations of static encryption mechanisms in dynamic and resource-constrained environments such as unmanned aerial systems (UAS). Traditional rule-based approaches offer deterministic performance and simplicity, but lack the flexibility to adapt to real-time changes in network bandwidth, CPU load, or security criticality of transmitted data. By integrating machine learning (ML) with predefined rules, the system leverages historical patterns and contextual awareness to refine encryption selection over time.
The rule-based phase acts as a stable default, ensuring compliance with baseline security policies and providing reliable encryption choices for known scenarios. In contrast, the ML-enhanced phase uses real-time telemetry, including features such as bandwidth availability, processing capacity, and mission-criticality flags, to predict the most dynamically appropriate encryption strategy. This dual-layered architecture not only improves decision speed and accuracy, but also future-proofs the system against emerging threats and mission variability. Furthermore, the ML model is trained using various operational scenarios to generalize across unseen network conditions, reducing overfitting and improving adaptability.
In practical deployment, this hybrid framework supports seamless transitions from conservative, rule-driven encryption to intelligent, data-driven adjustments, maintaining a balance between security strength, computational efficiency, and latency requirements. Over time, as the ML model continuously learns from more mission data, it can incrementally replace static logic, resulting in a self-optimizing encryption engine that supports real-time cybersecurity in complex environments.
Our hybrid model shown in Fig. 8 operates in two phases:
• Rule-Based Encryption Selection: Applies predefined encryption rules based on network conditions.
• Machine Learning-Based Selection: If the rule-based logic encounters uncertain or dynamic conditions, the ML model predicts the optimal encryption based on prior training.
To ensure secure and efficient data transmission in UAS, the ML-based approach is integrated to learn and optimize encryption selection over time. This approach ensures a safe transition from manual rule-based logic to autonomous, datadriven encryption selection.
The hybrid encryption algorithm operates as a sequential decision-making pipeline that dynamically selects the most suitable encryption method based on mission requirements and system resources. The process begins by initializing the encryption mode and classifying the data type as either missioncritical or routine. This classification directly influences the encryption sensitivity threshold, ensuring that mission-critical data are treated with enhanced security. Next, the system evaluates real-time telemetry on bandwidth and CPU load, essential parameters for balancing latency and security. The rule-based logic then provides an initial selection, drawing from predefined mappings between resource profiles and encryption schemes. For instance, AES-256 + RSA may be assigned to low-latency, high-security contexts, while AES-128 might be used in balanced scenarios. If this rule-based decision falls within a predefined confidence threshold, encryption is applied directly. However, if ambiguity or deviations from expected patterns are detected, the MLbased refinement module is triggered. This module evaluates historical behavior and feature vectors to predict the optimal encryption method with higher granularity.
The final selection is then executed, and the data is encrypted and transmitted securely. This algorithmic structure allows the system to combine the reliability of rules with the adaptability of machine learning, supporting both deterministic policy enforcement and intelligent responsiveness in dynamic UAS environments.
The hybrid encryption selection process follows the flow: Fig. 9. Hybrid Encryption Algorithmic Implementation Flow
• Data Classification: Identifies whether data is missioncritical or routine telemetry. Rule-Based Encryption Check: If a predefined rule applies, the system uses the appropriate encryption method.
• ML-Based Prediction: If conditions fall outside predefined rules or network fluctuations are detected, the ML model predicts the best encryption strategy.
• Encryption Execution: The selected encryption is applied, and data is securely transmitted.
A trained Random Forest Classifier was used to predict the optimal encryption method. The Random Forest model uses feature vectors composed of data size, bandwidth, and CPU load to classify encryption methods. Each decision tree evaluates threshold splits (e.g., if bandwidth ¡ 3 Mbps → prefer ChaCha20). The ensemble then votes on the optimal method. This transparency allows stakeholders to audit how decisions are made and provides a confidence score per prediction, adding interpretability as an essential aspect in security-sensitive applications where blind trust in black-box models is undesirable.
The ML-based model accuracy selection based on precision, recall, and F1 score was 100% based on the small sample data as shown below. The classification report is as follows:
precision recall f1-score support AES-256+RSA 1.00 1.00 1.00 3 accuracy 1.00 3 macro avg 1.00 1.00 1.00 3 weighted avg 1.00 1.00 1.00 3
1) Model Performance: The Random Forest classifier was evaluated using standard performance metrics, including precision, recall, and F1-score, for each encryption class (AES-128, AES-256 + RSA, ChaCha20). As shown in Table III, the ML-based model accuracy selection based on the metrics was 100%. This means that all three classes achieved perfect scores. Although these results are promising, their validity on larger datasets remains to be validated.
A confusion matrix and ROC analysis can also be incorporated in follow-up evaluations and studies.
TABLE III PERFORMANCE METRICS OF ML-BASED METHOD Encryption Method Precision Recall F1-Score Support AES-256 + RSA (High Security) 1.00 1.00 1.00 12 ChaCha20 (Low Latency) 1.00 1.00 1.00 33 AES-128 (Balanced) 1.00 1.00 1.00 55
A coded simulation was conducted to compare Rule-Based Encryption vs. ML-Predicted Encryption before proceeding to the hybrid model execution. The results are summarized in Table VI. We observed true matches selection from the models.
TABLE IV RULE-BASED VS. ML ENCRYPTION SELECTION # Data Size (KB) Bandwidth (Mbps) CPU Load (%) Rule-Based Encryption ML-Predicted Encryption Match? 1 650 7 35 AES-256 + RSA AES-256 + RSA True 2 200 2 85 ChaCha20 ChaCha20 True 3 450 4 60 AES-128 AES-128 True 4 700 8 40 AES-256 + RSA AES-256 + RSA True 5 300 3 70 AES-128 AES-128 True
The results in Table IV confirm that the ML-based model effectively replicates the decision logic of the rule-based approach across diverse network conditions. Each of the five scenarios presents a unique combination of data size, bandwidth, and CPU load, allowing the model to generalize encryption decisions reliably. The consistent matches across all test cases suggest that the ML model has successfully internalized the rule thresholds and applies them with high fidelity. Notably, in high-load and low-bandwidth conditions, the model consistently chose ChaCha20, aligning with low-latency priorities. Likewise, AES-256 + RSA was selected under moderate load and secure transmission conditions, and AES-128 was chosen in balanced scenarios, illustrating the hybrid system's nuanced handling of trade-offs. These results validate the model's ability to maintain encryption consistency and operational logic, even in the absence of rule-based control, making it suitable for real-time autonomous encryption decisions in UAS environments.
This further confirms that the selected random forest ML-based adaptive selection model as shown in Fig. 10
The results of our Hybrid Adaptive Encryption Algorithm for UAS demonstrate an adaptive approach to selecting the optimal encryption method based on data size, bandwidth, and CPU load as shown in Fig. 11 & Fig. 12.
For high-bandwidth and low-CPU load scenarios, we see that the algorithm prioritizes AES-256 + RSA for maximum security. In cases of limited bandwidth or high CPU load, ChaCha20 is chosen to optimize performance and reduce latency. When conditions are balanced, AES-128 is selected as a compromise between security and efficiency. This dynamic approach ensures secure and efficient data transmission while adapting to real-time system constraints.
To further validate the experiment of our proposed hybrid approach, a comparative analysis was conducted between Rule-Based, ML-Based, and the proposed Hybrid Model. Table V shows the comparison. The hybrid model executed an exact selection match from the rule-based and ML-based models. Fig. 12. Hybrid Encryption Model Execution and Selection Results
In this study, our hybrid adaptive model for decision selection demonstrated results that were closely aligned with both the Rulebased and ML-based models. While our findings provide valuable insights, it is important to acknowledge that they are based on a limited dataset used for this experiment. In practical UAS operations, where mission priorities, energy budgets, and communication links fluctuate rapidly, having an intelligently adaptive encryption mechanism is critical to both operational efficiency and mission success. Further validation with larger and more diverse datasets is necessary to fully assess the effectiveness of the model in live scenarios. For our case, the observation showed that this hybrid model successfully combines traditional rule-based logic with adaptive ML-based decision-making. The model can ensure security while allowing UAS encryption to evolve dynamically. It can also provides a structured transition from static encryption to fully automated AI-driven security.
The rule-based approach is reliable but static, meaning it does not adapt to new network threats or unpredictable conditions. Machine
TABLE V COMPARISON OF RULE-BASED, ML, AND HYBRID ENCRYPTION SELECTION Size (KB) BW (Mbps) CPU Load (%) Rule-Based Encryption ML-Predicted Encryption Hybrid Model Selection 650 7 35 AES-256+RSA AES-256+RSA AES-256+RSA 200 2 85 ChaCha20 ChaCha20 ChaCha20 450 4 60 AES-128 AES-128 AES-128 700 8 40 AES-256+RSA AES-256+RSA AES-256+RSA 300 3 70 AES-128 AES-128 AES-128 5th Intelligent Cybersecurity Conference (ICSC)
Learning offers dynamic adaptability, but it requires sufficient training data. The hybrid model ensures continuity and security while also allowing gradual ML integration. Additionally, our hybrid approach offers more benefits as it ensures encryption stability. ML only adjusts encryption when necessary, ensuring consistency while the hybrid model improves decision-Making, learns from previous encryption decisions, and refines selections over time. Therefore, the hybrid adaptive encryption model is recommended. It also reduces computational overhead and only triggers ML when network fluctuations or anomalies are detected. Furthermore, this approach supports modular updates, allowing new encryption algorithms or updated threat models to be integrated without overhauling the entire system.
Finally, although the model performed with 100% accuracy on the current dataset, it is important to note that the dataset was limited in scale. For further investigations, we intend to train the model on larger real-world datasets derived from live UAS telemetry to better evaluate generalization performance and prevent overfitting. We also aim to assess how the model responds under cyber-attack simulations, such as packet injection or data tampering, to evaluate its defensive robustness alongside adaptability. This will also allow stress testing under diverse mission scenarios and environmental variations.
The integration of real-world UAS networks into the training pipeline of Machine Learning models presents a promising avenue for enhancing encryption strategies. By leveraging live traffic data from operational UAS environments, ML models can be fine-tuned to better understand dynamic security threats, network fluctuations, and encryption performance in real-time scenarios.
Furthermore, the exploration of deep learning architectures, such as convolutional and recurrent neural networks (CNNs, RNNs), as well as transformer-based models, can significantly improve encryption decision-making by enabling adaptive security measures based on evolving network conditions. These models can enhance anomaly detection, intrusion prevention, and real-time cryptographic adjustments by learning complex patterns in UAS communications.
Additionally, the implementation of reinforcement learning (RL) can provide an autonomous approach to optimizing encryption strategies in real-time. By continuously analyzing network conditions, RL agents can dynamically select the most efficient encryption protocols, minimizing latency and computational overhead while maximizing security. Multi-agent RL frameworks can also be explored to allow multiple UAS within a fleet to collaboratively adjust their encryption settings based on shared security insights.
Future research can also investigate federated learning (FL) approaches to enable decentralized model training across distributed UAS nodes, preserving data privacy while continuously refining encryption strategies. Integrating blockchain-based key management systems with ML-driven encryption selection could further enhance the integrity, trust, and security of communication in UAS networks.
This study demonstrates that a hybrid encryption model, which combines rule-based and machine learning-based decision-making, offers the best of both methods. The hybrid method combines the predictability and reliability from rule-based logic and the adaptability and optimization from ML-based selection.
Authorized licensed use limited to: Embry-Riddle Aeronautical University. Downloaded on September 18,2025 at 20:04:21 UTC from IEEE Xplore. Restrictions apply.
5th Intelligent Cybersecurity Conference (ICSC)
Powered by TCPDF (www.tcpdf.org) Authorized licensed use limited to: Embry-Riddle Aeronautical University. Downloaded on September 18,2025 at 20:04:21 UTC from IEEE Xplore. Restrictions apply.