Document Type

Dissertation - Open Access

Award Date

2026

Degree Name

Doctor of Philosophy (PhD)

Department / School

Agricultural and Biosystems Engineering

First Advisor

Kasiviswanathan Muthukumarappa

Abstract

Smart Agriculture increasingly relies on image-based sensing, edge computing, and artificial intelligence to support data-driven decision-making in modern farming systems. While these technologies improve productivity and resource efficiency, they also introduce significant cybersecurity risks, particularly agricultural images transmitted and processed in real-time on resource-constrained edge devices. Unauthorized access, data manipulation, and interception of image data can lead to incorrect agronomic decisions and severe economic consequences. Despite this risk, efficient and practical cybersecurity solutions for real-time agricultural imaging remain limited. This study addresses these challenges by developing and evaluating encryption-based cybersecurity frameworks tailored for real-time agricultural image protection on edge computing platforms. First, a real-time full-frame image encryption framework based on the Advanced Encryption Standard (AES) with key lengths of 128 bits was implemented on NVIDIA Jetson Nano and Jetson Xavier NX devices using high-resolution soybean field imagery. The system was evaluated across multiple frame rates using statistical, differential, and performance metrics, including mean squared error, peak signal-to-noise ratio, pixel correlation, number of pixels changing rate, and unified average change intensity. Experimental results demonstrated strong resistance to statistical and differential attacks while revealing that the Jetson Xavier NX consistently outperformed the Jetson Nano, particularly at higher frame rates. To reduce computational overhead while maintaining cryptographic robustness, a selective encryption framework integrating a two-level two-dimensional discrete wavelet transform with AES encryption was then developed. By encrypting only the most information-rich wavelet coefficients, the proposed method significantly reduced processing latency while preserving strong security characteristics. Among the evaluated configurations, selective encryption combined with AES-192 achieved the most favorable balance between security strength and real-time performance, enabling stable operation at higher frame rates. Finally, an object-detection guided selective encryption system was introduced by integrating a deep learning-based YOLO object detection model with the encryption framework. This approach selectively protects semantically sensitive agricultural images, further improving efficiency without compromising data confidentiality. Experimental evaluation confirmed that object detection accuracy was preserved while maintaining strong encryption robustness. Overall, this study presents practical, edge-deployable cybersecurity solutions for protecting real-time agricultural image data. The proposed frameworks demonstrate that secure image encryption can be achieved on embedded edge devices without compromising performance or analytical capability, supporting the development of trustworthy and resilient smart agriculture systems.

Publisher

South Dakota State University

Download

Share

COinS
 

Rights Statement

In Copyright