of the failed large fruits, the suction device was unable to make a seal with the calyx because of the shape of the calyx.

Figure 9 shows the extracted fruit regions, calyx regions,and orientations of packed fruits in single-layer trays. The geometric features, like the orientation, the distance from the center of the hollow, and the rate of calyx region seemed to more closely resemble the results of manual packing.

Table 5 shows the data on the geometric features. The distance was converted from pixels to millimeters based on the scale factor of dpi of the pictures. Although the mean data were at almost same level in terms of distance, other features showed a significant difference between manual work and the packing unit. The orientations, in particular,varied widely for the packing unit and its rate for the calyx area was also significantly higher than that for manual work.

We believe that these errors were caused by three factors: first, the inaccuracy in the setting positions for placing fruits; second, inaccurate positioning of the singlelayer trays; and finally, the fruit’s posture of inclination before suction pickup. The first and second problems appeared to be soluble by tuning the packing unit. In respect of the inclination of the suctioned fruit,  Hayashi

et al. (2011) reported that the maximum angle of inclination tolerated by the suction device pickup was approximately 33°. To deal with this problem, a new machine vision system should be added to monitor the inclination of suctioned fruit and correct the geometric data for the

placing motion. As for the returnable tray, two fruits could not be placed with their apexes in the holes of the tray. It was also observed that these fruits were in an inclined posture. We believe that there are two approaches to solving this problem: the first would be to redesign the shape of the holes, such as by widening the chamfered corners, and the second would be to monitor the location of the apex of the fruit from the axis of the suction tube using new machine vision system. The packing unit should be able to locate the apex of the transported fruit at just above the center of the hole by correcting the placement position.Bruises were found on the apex part of two large fruits. We assumed that the apex parts had been pushed onto the cushion material of the returnable tray rather than being placed in the holes as intended. However, there were no bruises on the fruits in the single-layer trays. The damage rate was 2.7% with the packing unit.

PERFORMANCE OF THE AUTOMATIC PACKING SYSTEM

The task success rate of total fruits using the automatic packing system was 97.3% (table 6). The task success rates the supply unit and packing unit was 98.0% and 99.3%,respectively. In this test, the supply unit picked up all fruits from a harvesting container and safely put them on the conveyer. However, the spacing between fruits was sometimes short and the packing unit detected the risk of collision. The process of identifying the risk of collision was seen to operate correctly, whereas the spacemaintaining function of the supply unit did not work as intended. The packing unit on one occasion failed to pick up a large fruit and did not retry. Process time per fruit was 7.3 s overall due to the parallel operation of two units. The sorting and packing speed by hand is from approximately 5 to 8 s per fruit. Considering the feasibility of the automatic packing system, twice or more the working speed should be necessary. Meanwhile, it was observed that the supply unit was waiting while the packing unit was working. To reduce this standby time, we conclude that the automatic packing system should comprise a supply unit plus two packing units.论文网

Table 4. Process time and task success rate of the packing unit.

Table 5. Comparison of mean data of fruit posture packed in single-layer trays.

Table 6. Process time and task success rate of the automatic packing system.