In recent years, advancements in artificial intelligence and robotics have brought innovation to various industries. Specifically, real-time image classification technology plays a crucial role in autonomous vehicles, industrial robots, and medical robots. This paper presents a method for implementing an efficient real-time image classification system using Verilog HDL in a hardware environment and compares it with a software-based system. We implemented a fully connected layer using Verilog HDL and evaluated its performance using the MNIST dataset. The main advantage of the hardware implementation is its ability to increase computational speed through pipelining techniques and perform efficient operations with parallel processing of multiple computations. Experimental results show that the hardware-based approach significantly outperforms the software-based approach in terms of computational speed. This study confirms that hardware implementation has the potential to greatly enhance the efficiency of real-time image classification systems.