Table 7. Results of discrimination of fruit grade using machine vision system.

ACCURACY OF GRADING FUNCTION

Table 7 shows the result of discrimination of fruit grades by the machine vision system. The matching rate between human and machine vision system for each size was 85.2%,70.8%, and 59.0%, respectively. Fruit weight histograms of the graded fruits by a farmer and the automatic packing system are shown in figure 10. It was considered that the matching rates became relatively low because the results of grading by a farmer varied widely. The weight estimation error ranged from -1.9 to +2.0 g, and standard deviation was 0.7 g for small fruits and 0.8 g for the other sizes. We believe that the weight estimation errors were chiefly caused by the three-dimensional shape features such as the difference between short and long diameters. The density of the fruit and the shape of the calyx also appeared to affect the measurement accuracy. In cases of other agricultural products such as waxed apples, potatoes, and tomatoes, a machine vision system that consists of two or more cameras could improve the accuracy of mass estimation for automatic grading (Kurita and Kondo, 2006). Compared to a conventional sorting device with a weight scale, the accuracy of weight estimation with the machine vision system for strawberries appeared to be inferior. If the automatic packing system is enhanced to feature a more precise grading function, the installation of a weight scale might be a practical solution.

CONCLUSIONS

An automatic packing system for strawberries was constructed to pack fruits methodically in

returnable trays or single-layer trays. The system was comprised of a supply unit and a packing unit. Each unit had machine vision system, a manipulator, and a suction hand that operated simultaneously. The supply unit detected a fruit in a random orientation without overlap in a harvesting container, picked it up, and placed it gently on the fruit conveyer. The process time per fruit and the task success rate was 4.5 s and 99.2%, respectively. There were bruises on 4.1% of tested fruits when using the supply unit. Improvement of the machine vision system of the supply unit was thought to be a solution for reducing the bruises. The packing unit had the role of locating the orientation and position of the supplied fruit on the conveyer, grading it according to its estimated weight, picking it up, and placing it on a shipping tray. The process time per fruit and the task success rate was 6.3 s and 98.4%, respectively.However, the fruit postures in the single-layer tray differed significantly from the same job done manually. There were bruises on 2.7% of tested fruits with the packing unit. For the uniformity of packed fruit posture and the reduction of bruises, monitoring the inclination of sucked fruit and the correct geometric data for the placing motion should be needed. Process time per fruit and task success rate of the automatic packing system was 7.3 s and 97.3%, respectively. It was revealed that the reduction of stand-by time should be required. The matching rate of fruit-grading results between a farmer and machine vision system varied from 59.0% to 85.2% according to grade, while the estimation error of fruit weight ranged from -1.9 to +2.0 g. If the automatic packing system is enhanced to feature a more precise grading function, the installation of a weight scale might be a practical solution.

摘要为了减少日本草莓生产的劳动力，我们开发了一种用于草莓的自动包装系统，将果实有条不紊地包装在可回收的托盘或单层托盘中。该系统包括供应单元和包装单元，每个单元具有机器视觉系统，操纵器和并联操作的抽吸手。 供料单元负责在收获容器中以随机姿势检测果实，拾取并轻轻地将其放置在输送机上。包装单元然后将供应的水果的方向和位置定位在输送机上，根据其估计的重量对其进行分级，拾取它并将其放置在运输托盘中。 生产区的性能测试显示，自动包装系统的任务成功率为97.3%，每个处理时间为7.3s。然而，根据等级，手动分级和机器视觉系统的比例显示从59.0％到85.2％的变化，而果实重量的估计误差范围为-1.9g至+ 2.0g。我们还对各个单位进行了性能测试。供应单位受损果实比例为4.1％，包装单位为2.7%。分析了包装水果的姿势，显示了包装单位和手动操作结果之间的差异。需要进一步研究，特别是减少水果上的瘀伤和包装果实的姿势的均匀性。