Development of Potato Yield Monitoring System Using Machine Vision

The study is aimed at investigating the feasibility of using machine vision for potato yield monitoring systems. Therefore, it is necessary to develop experimental-scale hardware and necessary algorithms for monitoring systems and to evaluate the system performance. The experimental apparatus consisted of a digital camera with an auto-iris lens, a gimbal system, a camera trigger device on a driving wheel, RTK-GPS, and a laptop computer with MATLAB installed. In a potato farm, the images of potatoes taken out of soil and scattered on the soil surface were acquired at harvest time, and the estimated mass of the potatoes was obtained through image processing and a mass estimation algorithm. Actual masses of the potatoes were measured and recorded for each specific divided region in the field to evaluate system performance. The algorithm for distinguishing between potato and soil noise worked well. It required 103.4 ms to complete the image processing on a single-frame image. The integrated algorithm required 344.0 ms to output the results, such as the estimated mass with GPS location information, from an acquired image. The system performance evaluation results showed that the yield estimation error was 268.60 g and 15.33% in root mean square deviation (RMSD) and percentage error, respectively. The monitoring system used in this study is highly significant in the sense that it is possible to make a relative comparison of variations in the field.