A Real-time Citrus Segmentation and Detection System using Mask R-CNN

• Published in: Journal of Digital Contents Society Vol. 19; no. 12; pp. 2385 - 2391
• Main Authors: Kim, Jin-Won, Lee, Malrey
• Format: Journal Article
• Language: English
• Published: 한국디지털콘텐츠학회 31.12.2018
• Subjects: 컴퓨터학; 빠른 학습; Real-time performance; 과일 검출; Visual fruit detection; 실시간 성능; 자동 수확; 심층 콘볼루션 신경망; Deep convolutional neural network; Rapid training; Auto-harvesting
• ISSN: 1598-2009; 2287-738X
• DOI: 10.9728/dcs.2018.19.12.2385

In this paper, 200 photograph of citrus were collected and converted to 800x800. The areas of each citrus in the photograph were mask-labeled and stored in JSON format to generate a data set. The latest algorithm, Mask R-CNN, I constructed a reliable system to detect and divide citrus fruits. In order to solve the over-fitting problem due to small data sets, the data augmentation was used and the detection performance was as high as 0.97 with small data sets. In order to meet the farmer's practical needs, I plan to develop a platform that can take photographs, label them with a mask first, and then train them immediately after doing additional mask labeling work. KCI Citation Count: 0