Tomato volume and mass estimation using computer vision and machine learning algorithms: Cherry tomato model
- Resource Type
- Authors
- Liu Chao; Luke Nyalala; Cedric Okinda; Kunjie Chen; Khurram Yousaf; Nelson Makange; Qi Chao; Innocent Nyalala
- Source
- Journal of Food Engineering. 263:288-298
- Subject
- Mean squared error
Machine vision
Image processing
Machine learning
computer.software_genre
03 medical and health sciences
0404 agricultural biotechnology
0302 clinical medicine
Cherry tomato
Computer vision
Power function
Mathematics
biology
business.industry
Sorting
04 agricultural and veterinary sciences
biology.organism_classification
040401 food science
030221 ophthalmology & optometry
Artificial intelligence
business
Algorithm
computer
Predictive modelling
Food Science
Volume (compression)
- Language
- ISSN
- 0260-8774
A prediction method of mass and volume of cherry tomato based on a computer vision system and machine learning algorithms were introduced in this study. The relation between tomato mass and volume was established as M = 1.312 V 0.9551 , and was used to estimate mass on a test dataset at an R 2 of 0.9824 and RMSE of 15.84g. Depth images of tomatoes at different orientations were acquired and features extracted by image processing techniques. Five regression prediction models based on 2D and 3D image features were developed. The RBF-SVM outperformed all explored models with an accuracy of 0.9706 (only 2D features) and 0.9694 (all features) in mass and volume estimation respectively. The model predicted mass or volume can then be applied to the established mass-volume power function. This introduced system can be applied as a non-destructive, accurate and consistent technique to in-line sorting and grading of cherry tomatoes based on mass, volume or density.