EDGE-ENABLED HYBRID DEEP LEARNING FRAMEWORK FOR IOT-BASED PRECISION SMART AGRICULTURE: DESIGN AND PERFORMANCE EVALUATION

Abstract

Precision agriculture can utilize the power of data-driven technologies, including the Internet of Things and Deep Learning, in order to streamline agricultural activities, maximize agricultural output, and save resources. Nevertheless, most of the current systems rely on cloud processing and set rules, which results in high latency and predictive capability. This paper suggests an Edge-Enabled Hybrid Deep Learning Framework that incorporates the Long Short-Term Memory networks to time-series soil moisture prediction and Convolutional Neural Networks to crop disease detection into an iot model. The inference of deep learning is implemented on the edge to minimize latency and bandwidth consumption and allow making decisions on irrigation and plant health in real-time. The analysis of performance on publicly available datasets shows that it has a higher accuracy in prediction and response time than conventional machine learning models. The framework measures the latency reduction and comparative consumption energy between edge and cloud deployments.