Machine vision-based tomato plug tray missed seeding detection and empty cell replanting

Abstract

The current tomato plug tray sowing methods are plagued by the issue of missed sowing. Consequently, the detection of missed sowing and empty cells replenishment of tomato plug trays after initial sowing is an important prerequisite to realizing precision sowing of plug trays and improving the quality of plug tray seedlings. This study proposed a machine vision-based information perception method for tomato plug trays to detect missed seeds and replenish empty cells. First, based on a YOLOv5x network model, deep learning training was employed to obtain a detection model for identifying tomato seeds in complex backgrounds of plug trays. Thereafter, the location of the tomato seeds was compared with the coordinate value of the edge of the hole, and the missed holes were detected. Based on the missed cells detection results, the empty cell coordinates were converted into replanting coordinates, and the plug tray conveyor and the replanting suction needle are controlled separately considering the replanting row information to replant the missed cavities row by row. Further, the tomato plug tray missed seeding detection and replenishment device was built and tested at a seedling greenhouse. The results indicated an average detection accuracy of 92.84 % with average detection time per tray of 13.475 s. The success rate of replanting on trays was determined to be 91.7 % with 5 %-20 % misses. In addition, the productivity of the tray was approximately 42.4 trays/h. Thus, the proposed method of detecting and replanting misses in tomato trays can be used to improve the performance of tomato seeding. Moreover, it can be used as a reference to improve the precision and efficiency of tomato seeding in plug trays.