FPGA-based Features Extraction Sensor for Lettuce Crop

Abstract

A method for extracting lettuce phenotypic features using an ARTY A7-35T FPGA platform is proposed. Image acquisition is done by interfacing an OV7670 CMOS camera with FPGA and saving image data on DDR3 memory. The image processing techniques firstly involve color model conversion for saturation enhancement. Then, binarization is done as an initial step in background discrimination, using a threshold value on the green channel of image. To construct a solid figure for foreground, morphological transformations are implemented. Then, pixel count of foreground white pixels are used as an argument in the computation of lettuce canopy area. The FPGA implementation consumes 85.89% of total LUTs resources of the chosen FPGA board with errors as low as 1.28% on the computed canopy area value compared with a MATLAB benchmark. Power consumption reached up to 1.73W, with a total calculated latency of 596.51 ms from image acquisition to canopy area value.