Accelerating Deep Neural Network-Based Image Analysis Using Reconfigurable Architecture

This study explores the utilization of reconfigurable architectures, specifically field-programmable gate arrays (FPGAs), for accelerating image classification tasks based on deep neural networks (DNNs). The COVID-19 Radiography Database, consisting of COVID-19 chest X-ray images, is analyzed using three different DNN classification model architectures. The trained model weights are compiled for hardware-specific execution on an FPGA board, taking advantage of the parallel processing capabilities and hardware acceleration offered by the DPU architecture. Performance in terms of the accuracy of the model is evaluated along with inference time and resource utilization. The experimental results demonstrate significant improvements in speed and energy efficiency through the integration of software and hardware components. The findings provide valuable insights into the potential of Field Programmable Gate Array (FPGAs) and Deep Learning Processing Units (DPUs) in enhancing image analysis tasks based on deep neural networks, paving the way for future advancements in this field.