The invention provides a tobacco leaf grading method and device based on a FPGA (field programmable gate array). The device comprises an interface board FPGA and a processing board FPGA which are mutually connected through double rows of insertion pins, wherein the interface board FPGA is connected with a CCD (charge coupled device) through a serial communication chip interface; the interface board FPGA sends an instruction to collect a frame image; the collected image is stored into a first SRAM (static random access memory) cache; the interface board FPGA reads the collected tobacco image to perform background division and performs denoising processing; then, set image features are extracted; the interface board FPGA stores the extracted image features into a second SRAM cache; then, the processing board FPGA processes the feature information of tobacco leaves according to a built grading model algorithm to obtain a color separation grading result; the color separation grading result and the control information are sent into the interface board FPGA; the interface board FPGA displays the collected tobacco leaf image and the corresponding color separation grading result on a VGA (video graphics array) and broadcasts the collected tobacco leaf image and the corresponding color separation grading result through a voice module. The two FPGA chips are used as main control chips; in the speed aspect, the processing speed on the tobacco image is greatly accelerated.