Field performance of parallel yield sensors for measuring individual weights of onion bulb

Field performance of parallel yield sensors for measuring individual weights of onion bulb

Yang, Y.P. Book author
Wu, G.J. Book author
Chiu, Y.C. Book author

Yi-Lan Department of Biomechatrioic Engineering, National Ilan University Taiwan

2014

p.642-647

EAM0576

Individual weight measurements of tubers, bulbs, and fruits on the harvester will feed back to enhance their market value and induce progress in Precision Agriculture. On an onion picker, 10 impact-based parallel yield sensors were mounted, and experiments were carried out in 2013 cropping season. Each sensor consisted of a cantilever load cell (a 38 mm × 208 mm × 4.5 mm hard steel plate with 4 strain gauges) and a sponge rubber cushion (10 mm in thickness), and was installed under the end of the lifting conveyor. Calibration was first carried out indoor to relate the impulses received by the sensors to the individual weights of the bulbs, where 23 bulbs (43 g to 306 g) were used. The calibration was based on a nonlinear model: w = a p ^ b, where w is the weight of the bulb, p the impulse received by the sensors, and a and b are the parameters; it showed the standard error of 26.3 g and coefficient of determination (R-square) of 0.84. As the validation, field measurements were carried out, where the bulbs were harvested in 31 containers (ca. 20 kg each). After the harvest, all the bulbs in each container were manually weighed, resulted in total of 2090 bulbs ranging from 16 g to 551 g, and they were classified into 9 classes (3S to XL) according to a Japanese market standard. For each container, the Kolmogorov-Smirnov test was performed of the distribution in the actual and the estimated weights at the significance level of 5%. The test showed, however, that the similarity in the actual and estimated distributions was rejected for 3 out of 31 containers. Therefore, re-calibrations of the sensors were attempted using the bulbs in arbitral 5 containers harvested. The same non-linear model was used, and the parameters were optimized to obtain similarity between the actual and the estimated weight distributions. Consequently, apparent discrepancy was eliminated and the similarity in the weight distributions was accepted for 25 out of the remaining 26 containers. Additionally, the harvester ran in a 10 m × 80 m field and a conventional yield map and a mode map was generated, based on the re-calibration.

International symposium on machinery and mechatronics for agricultural and biosystems engineering; Proceedings of the

Research > Published Articles