Publication

Abstract : A Convolutional Neural Network (CNN) are a class of artificial neural networks specifically designed to process data with a grid-like topology, such as images, making them well-suited for tasks like image recognition and classification, object detection, and speech recognition. However, their current software implementations leave much to be desired regarding energy efficiency, speed, performance and scalability. Hardware-based implementation of CNNs has several significant advantages over software, including faster data processing due to the parallel execution of hardware-based FPGAs and ASICs, which are necessary for real-time applications. They are more energy-efficient and have consistent, predictable performance. In this paper, we present the implementation of CNN on Verilog. We implemented a highly optimized CNN regression model architecture for object detection having an accuracy of 97%. The entire design was made on Verilog, allowing easy transferability to both FPGA and ASIC platforms. The proposed work is compared to the current standard for software implementations – Google Colab. The results obtained show considerable speed up and improved performance.