according to the farmer’s grading results. Measurements of bruises were conducted in the same

way as for the supply unit.

Performance of the Automatic Packing System

To evaluate the efficiency and the accuracy of the total system, comprising the supply unit and the packing unit,we conducted the continuous automated tasks from picking up the fruit from the harvesting container and putting it on the conveyer at the correct intervals, to picking it up and placing it in shipping trays. The process time and task success rate of the system were obtained and their breakdown for inpidual units were also measured.

Accuracy of Grading Task

To grade strawberry fruits by size, we prepared a regression line in advance which was based on the data for 54 ‘Sagahonoka’fruits ranging in weight from 8 to 16 g:y = 2.213×10 -3 x– 3.626, r2 = 0.87, where x is the number of pixels in a fruit region, y is estimated fruit weight (g),

and r2 is the determination coefficient. For grading the fruit,the threshold weights of the fruit were set at 10 and 14 g,i.e., small fruits weighed less than 10 g, medium fruits were 10 to 14 g, and large fruits weighed more than 14 g. The grading test was conducted in part of the performance test of the automatic packing system. The automatic packing system sorted the fruit into the three classes. We recorded the grading results by the system through the packing operation and compared them with the manual process carried out by a farmer.

Table 3. Process time, task and suction success rate of the supply unit.

Figure 9. Fruit posture in a single-layer tray packed (a) by hand and

(b) by the packing unit.

RESULTS AND DISCUSSION PERFORMANCE OF THE SUPPLY UNIT

Table 3 shows the results of the performance test of the supply unit. Process time per fruit was 4.5 to 4.6 s. The machine vision system detected all the fruits; however, one medium-sized fruit failed to be picked up with two suction attempts, giving a total task success rate of 99.2%. Meanwhile, the suction success rate of medium-sized fruits dropped to 90.4% and that of total fruits was 95.5%. The number of fruits which needed two suction attempts was six, or 4.7% of successfully handled fruits. It was observed that the unit could definitely handle a fruit if the position information on the targeted fruit was accurate. We thus concluded that it was chiefly machine vision system that was responsible for suction failures. Bruises were found on three of the small fruits. Two were relatively small; however, an indentation was found on the equatorial segment of the remaining fruit. Since we observed no bruises on the other tested fruits, the damage rate was 4.1%. It appeared that the fruits were damaged by the stress concentration of the suction tube which was itself caused by the variety in hardness and shape of the strawberry fruit that could not be counteracted by the cushioning material. Although this bruising problem might be solved if the suction force were reduced, the robustness against error of machine vision system would also be reduced, and the task success rate would drop accordingly. It became clear that improved machine vision system would be needed to reduce bruising. Current machine vision system cannot distinguish reflective parts from uncolored parts of strawberry fruit. In tomatoes, the reflective part should be the highest part of the fruit as Tillett et al. (1995) reported, but this is not always the case in strawberries. To improve the machine vision system, a 3D sensor should be applied to the supply unit to detect the highest part of a fruit in harvesting container more accurately.

PERFORMANCE OF THE PACKING UNIT

The success rates of the packing unit are shown in table 4. Process time per fruit varied from

5.7 to 6.4 s. The times needed for large fruits were relatively short because the placing motion for the returnable tray was simpler than that for the single-layer tray. The overall task success rate was 98.4%. One small fruit and two large fruits failed to be suctioned by their calyx. Due to the calyx leaves of the failed small fruit being relatively large and separate from the fruit body, the machine vision system was unable to locate an accurate suction point for the fruit. In the case