Research and Implementation of Extended Instruction Set for Smart Industrial Vision

With the rapid development of artificial intelligence technology, the demand for efficient image processing in automated inspection and quality control within intelligent industrial vision is increasing. However, existing general-purpose processors often struggle to meet the industrial-grade performance requirements for handling complex image-processing tasks. To address this challenge, this paper proposes a set of extended instruction sets based on the RISC-V architecture to accelerate computations for median filtering, gradient squared computation, threshold comparison, and image greyscaling. The consumption of decoding logic resources is reduced by using one-hot encoding for Funct7. A dedicated Image Processing Unit (IPU) is also designed to accelerate image processing computations further. The experiments were conducted on a Xilinx Artix-7 XC7A35T FPGA platform, and the results demonstrate that, compared to processors using the standard RISC-V integer instruction set, the processing cycles of the four designed extended instructions are reduced by 73% to 85%. Moreover, the overall image processing performance is improved by four times, significantly enhancing the system's execution efficiency.