S-Space College of Agriculture and Life Sciences (농업생명과학대학) Dept. of Biosystems and Biomaterials Science and Engineering (바이오시스템·소재학부) Theses (Master's Degree_바이오시스템·소재학부)
On-the-go Embedded System for Spatial Mapping of Lettuce Fresh Weight in Plant Factory
식물공장 내 상추의 생체중 공간 매핑을 위한 이동형 임베디드 시스템 개발
- 농업생명과학대학 바이오시스템·소재학부(바이오시스템공학)
- Issue Date
- 서울대학교 대학원
- 학위논문 (석사)-- 서울대학교 대학원 : 농업생명과학대학 바이오시스템·소재학부(바이오시스템공학), 2018. 8. 김학진 .
- Real-time monitoring of crop growth parameters in plant factory can provide useful information about accurate assessment of their growth status for precision crop management. Plant weight is one of the most important biophysical properties used to determine the optimum time for harvesting. Conventional plant weight measurements are destructive and laborious. An on-the-go image processing system consisting of image acquisition and weight estimation was developed to generate a fresh weight map of lettuces grown in hydroponic solutions. Key technologies developed in this study include a robust image preprocessing technique that separates individual lettuce images in the presence of overlapping leaves and a real-time image processing method that estimates their fresh weights. Images were captured with a low cost web camera and processed using a myRIO-based embedded controller. The camera and embedded system moved along an XY axis frame above a plant growing bed (0.94 x 1.8m) using two stepping motors and linear actuators. The image preprocessing algorithm consisted of three main subroutines, i.e., image segmentation, target plant recognition and overlapping leaf separation. For the image segmentation, the S channel of the HSV color space and Otsus threshold were used to separate the plants from the background. The target plant was identified based on the mass center of the region of interest. The overlapping leaves were removed using an iterative erosion method. The plant weight was calculated by counting the number of pixels automatically. The accuracy of the fresh weight estimation by the system showed a highly linear regression with a slope of 1 and a determination coefficient (R2) of 0.9. The results showed that it was possible to measure the plant fresh weight of each lettuce in real time. And the overlapping leaf problem can be solved until the harvesting of lettuces. It was expected the developed system would be a useful tool for mapping fresh weights of lettuces grown in a plant factory.