Abstract:Selection and screening of soybean seeds was an important link in soybean seeds processing. At present, manual work and mechanical principle were widely used in domestic selection and screening of soybean seeds, which were featured by high cost, great labor intensity and low efficiency and precision. In recent years, with the research on machine vision technology deeper, the machine vision was more and more widely applied to recognition and detection of agricultural products. A total design scheme of embedded soybean seeds visual classification system was proposed based on DSP and ARM. The working principle of the device, hardware configuration, software system and placement test were introduced. A soybean seeds selection algorithm was designed, and statistic on parameters was made, extraction of soybean seeds boundary and regional drawing characteristics was found out, and roundness and smoothness were set as primary basis of selection. DSP-ARM dual-processor architecture processor with DaVinci technology TMS320DM6437 (DSP) and TMS320DM355 (ARM) was used as the core processing unit. In this system, a realtime process to the soybean seeds picture captured by camera was taken by using DaVinci technology TMS320DM6437, and the processed result was obtained by using TMS320DM355, which achieved the intelligent grading of soybean seeds. Image processing algorithm was designed firstly, statistical approach was utilized to distill boundary characteristics and regional characteristics of soybean seeds, and then the grading feature was determined. Visual grading of soybean seeds was achieved by embedded operating system. Four varieties of soybean (Dongnong 42, Dongnong 89836, Dongnong L13 and Dongnong 44) of 2000 grains each were taken as test samples to retest the device. The precision of selection and screening can reach 95%.
