在线式玉米单粒种子检测分选装置设计与试验

针对农业生产中种子精选的需求，设计了在线式单粒种子检测分选装置，实现流水线式种子上料、检测和分选。该装置由上料装置、检测单元、分选单元和控制系统组成。上料装置通过两级振动实现籽粒的平铺，配合传输带完成籽粒的单粒化。检测单元由高速工业相机实时获取种子图像，并传送至上位机检测分析。控制系统根据检测结果和种子在图像中的位置，控制分选单元完成分选。利用搭建的装置采集了1 200粒正常种子、1 200粒霉变种子和1 200粒破损种子的图像，使用HALCON软件提取了单粒种子的18个颜色和12个形态特征，通过偏最小二乘判别分析法进行判别分析，分别构建了种子霉变和破损的检测模型，并利用搭建的装置和模型进行了验证试验。试验结果表明：在线式单粒种子检测分选装置分选速率大于300粒/min;其中霉变种子的分选准确率高于95%,破损种子分选的准确率高于89%。

Design and Experiment of Online Maize Single Seed Detection and Sorting Device

With the development of single seed sowing and precision seeding technology in China,higher requirements are put forward for the quality of single seed. In response to the current demand for fine seed selection in agricultural production,an online single seed detection and sorting device was designed. The device consisted of a feeding device, a detection unit, a sorting unit and a control system. The feeding device was composed of two sets of linear vibration devices, which can realize the flattening of the grains through two-stage vibration, and cooperate with the conveyor belt to complete the single granulation of the grains. The detection unit obtained the seed image in real time by the high-speed industrial camera, and transmitted it to the upper computer for detection and analysis. The sorting unit was made of sorting components and air compressor, which was used to remove the identified damaged or moldy grains. The control system controlled the sorting unit to complete the sorting according to the detection result and the position of the seed in the image. Furthermore, images of 3600 corn seeds (1200 normal seeds,1200 moldy seeds, and 1200 damaged seeds) were collected by using the built device, and image processing algorithms were used to obtain the 18 color and 12 morphological characteristics of a single seed, and the partial least squares discrimination analysis method (PLSDA) was used for discriminant analysis, and the detection models of moldy and damaged seeds were constructed respectively. Then the online verification experiment was carried out by using the built device and model. The results showed that the sorting rate of the device was greater than 300 seeds/min; the sorting accuracy of the mildew model was higher than 95%, and the sorting accuracy of the damaged model was higher than 89%. The device can realize the full automation of corn seeds from feeding to sorting, and can detect and sort moldy and damaged corn seeds in real time.