Deep Learning Applications in Embedded Systems and IoT Devices
G Praveen, Ashok Walikar, Syed Riyaz Ahammed, Yashas Suresh, Hitesh Parihar, Vasupalli Manoj · 2025
In this paper, we have presented a detailed analysis of deep learning-based systems on different embedded platforms and IoT devices. We have tested the performance of CNN, LSTM, and DBN models on three different platforms: Raspberry Pi 4, Nvidia Jetson Nano, and ARM Cortex-M4. Our analysis in terms of accuracy, latency, power, and memory use of these models has enabled us to assess the suitability of these methods in resource-constrained environments. Our test concludes that Nvidia Jetson has the highest accuracy of 98.3% at a very low latency rate, making it more appropriate for realtime high accuracy-demanding applications. On the other hand, the Raspberry Pi 4 has an accuracy of 98.1 % and a modestly latency performance level, making it best for a balance of performance and power cost-effective systems. ARM Cortex-M4 has a high latency performance and poor accuracy based on its very low computational and memory function. It is best applicable systems where power consumption is a sensitive issue. Due to the implications of our findings, we recommend the use of optimization mechanisms to facilitate the running of such complex models within the microcontroller-based systems. Therefore, this study provides an essential insight for the deployment of deep learning-based systems on embedded platforms and IoT. It will be our next step to optimize our models on the microcontroller platform. Similarly, the complexity of the network will be increased to obtain baise suitability.