Real-Time Morphological Measurement of Oriental Melon Fruit Through Multi-Depth Camera Three-Dimensional Reconstruction

• Published in: Food and bioprocess technology Vol. 17; no. 12; pp. 5038 - 5052
• Main Authors: Hong, Suk-Ju, Kim, Jinse, Lee, Ahyeong
• Format: Journal Article
• Language: English
• Published: New York Springer US 01-12-2024; Springer Nature B.V
• Subjects: Agriculture; Algorithms; Biotechnology; Cameras; Chemistry; Chemistry and Materials Science; Chemistry/Food Science; Feature extraction; Food Science; Fruits; Image quality; Image reconstruction; Imaging techniques; Melons; Morphology; Parameters; Phenotyping; Prediction models; Real time; Three dimensional imaging; Time measurement; Oriental melon; 3D reconstruction; Point cloud; Depth camera; Grading
• ISSN: 1935-5130; 1935-5149
• DOI: 10.1007/s11947-024-03435-8

Morphological features of fruit, such as size and shape, are essential in determining fruit quality. Given the limitations in accurately measuring precise morphological features using solely two-dimensional (2D) images, studies utilizing three-dimensional (3D) imaging techniques for measuring fruit morphology have been conducted. However, because of the time-consuming processes involved, measuring and processing 3D images in real time has thus far been impossible. Therefore, this study aimed to measure 3D images and extract the morphological features of fruits in real time. A measurement system with multiple RGB-D cameras was developed to enable real-time measurements by coordinate calibration among the cameras. Algorithms for real-time extraction of morphological features specific to oriental melon fruits were also developed. The prediction performances for the length, volume, and density of oriental melons showed determination coefficients of 0.9676, 0.9975, and 0.9057 and root-mean-squared errors of 2.08 mm, 3.77 cm
3
, and 10.73 kg/m
3
, respectively. In addition, predictive modeling was performed for their morphological grades by using parameters based on 3D morphology. The reference grade was determined by skilled workers at the processing center according to their produce classifying standards. The parameters were analyzed against their morphological grades, and the predictive model showed an accuracy of over 94%. The developed system and algorithms had a processing time of 40.28 ms for measuring and processing 3D images with an i5-16300KF CPU and 32 GB of RAM, indicating their potential application in phenotyping and as fruit-sorting machines.