Adaptive Edge Intelligence Framework for Real-Time Multimodal Data Processing in Resource-Constrained IoT Networks

Abstract

The rapid expansion of Internet of Things networks has resulted in the continuous generation of heterogeneous data from cameras, microphones, environmental sensors, wearable devices, and industrial monitoring units. Processing these data streams exclusively through cloud platforms can introduce communication delays, increase network traffic, and create additional concerns related to privacy and service availability. These limitations become more significant when IoT devices operate with restricted computing capability, memory capacity, storage, and energy supply. This research proposes an Adaptive Edge Intelligence Framework for real-time processing of multimodal IoT data under resource-constrained conditions. The proposed framework integrates lightweight modality-specific processing, adaptive feature extraction, multimodal data fusion, resource-aware model selection, and dynamic edge-cloud task allocation. Instead of transmitting all collected data to a remote server, the framework determines the appropriate processing location according to computational load, available memory, energy status, network conditions, and application latency requirements. A hierarchical processing mechanism performs initial filtering and feature extraction close to the data source, followed by selective fusion and inference at an edge gateway. The framework also supports dynamic adjustment of processing complexity when device resources or network conditions change. The design is intended to reduce unnecessary data transmission while maintaining reliable real-time inference across heterogeneous IoT applications. The proposed approach provides an architectural foundation for deploying adaptive multimodal intelligence on low-power IoT networks and establishes a basis for evaluating trade-offs among inference accuracy, latency, communication overhead, memory consumption, and energy usage.