Abstract—This study focused on grading the quality of fruit in packaging plants. Fruit grading requires a body of standard data, which is difficult to achieve in practice. In most studies, relevant data are established solely by the researcher during the course of the research. We mounted cameras at fixed locations on machines to collect image and weight data, allowing farmers to create data feeds without disrupting their work. The system consists of a database of 400 images that were manually labeled and trained on a deep learning network architecture. During the training process, 70% of the images in the database were randomly selected for training, and the other 30% were used for verification to ensure that the training process did not over-learn, as over-learning leads to a decrease in the recognition rate. After the position of the fruit in the image was detected through deep learning, the foreground and background were separated, the information about the fruit was extracted, and the total number of pixels was calculated. Automatic measurement was achieved by converting pixels to millimeters using standards in the environment. The detection rate of the proposed system was over 98%. Using 50 manual measurements of fruit size and automatic detection results for error analysis, the diameter error value was 15.3 mm and the length error value was 14.45 mm.

Keywords—automatic measurement, automatic detection, Intelligent Labor Saving

Cite: Ying-Tzy Jou, Yun-Yun Shih, and Chia-Ying Chang, "Measuring Fruit Size via Deep Learning Using a Single Camera," Journal of Advanced Agricultural Technologies, Vol. 11, No. 2, pp. 28-32, 2024.

Copyright © 2024 by the authors. This is an open access article distributed under the Creative Commons Attribution License (CC BY-NC-ND 4.0), which permits use, distribution and reproduction in any medium, provided that the article is properly cited, the use is non-commercial and no modifications or adaptations are made.