Spectrum sensing is a critical task in dynamic spectrum access, where accurate identification of occupied and vacant spectrum bands enhances wireless communication efficiency. This paper proposes a quantized semantic segmentation model for spectrum sensing, optimized for FPGA deployment to achieve real-time processing with minimal resource usage. The model is trained and quantized to 16-bit and 8-bit precision, then deployed using the FINN framework for efficient FPGA acceleration. Experimental results show that our approach maintains high accuracy exceeding 90% in controlled scenarios and over 70% on real-world data, while significantly reducing latency and resource consumption. These results demonstrate the feasibility of our approach for real-time spectrum monitoring in energy-constrained environments.