Image Processing for Smart Agriculture Applications Using Cloud-Fog Computing

Area of Science:

• Agricultural Technology
• Computer Science
• Artificial Intelligence

Background:

• The Internet of Things (IoT) generates vast data, necessitating advanced analytical solutions, particularly in smart agriculture for crop monitoring.
• Continuous monitoring of crop growth aids in timely interventions for disease, weed, and pest control, boosting agricultural productivity and sustainability.
• Image analysis using Convolutional Neural Networks (CNNs) offers significant potential for enhancing decision-making systems in smart agriculture.

Purpose of the Study:

• To develop a deep learning model for image classification optimized for resource-constrained Fog computing devices.
• To enable early problem detection and optimize resource management in smart agriculture through efficient image processing.
• To reduce agricultural operating costs and manual labor by leveraging edge and fog computing for data processing.

Main Methods:

• Implementation of a Fog computing architecture connecting Cloud and Edge devices for data processing.
• Development and optimization of a deep learning model for image classification suitable for hardware-limited devices.
• Testing a tomato disease classification model on Field-Programmable Gate Arrays (FPGAs) to evaluate performance trade-offs.

Main Results:

• The proposed solution effectively off-loads data processing to Edge and Fog devices, improving system responsiveness and reliability.
• Significant reductions in data transmission and storage costs were achieved.
• The optimized model for FPGA execution showed a minimal decrease in test accuracy (0.83%) for tomato disease classification, maintaining high performance (95.46%).

Conclusions:

• The developed deep learning model is adaptable for implementation on resource-limited Fog computing devices, enhancing smart agriculture applications.
• Optimizing image processing at the edge significantly improves system efficiency, reduces costs, and increases overall reliability and security.
• The approach demonstrates a viable method for balancing model size and accuracy, crucial for deploying AI in edge environments for agriculture.