SAAIL Lab Publications
J. Food Engineering · 2019 5.3 178+ Citations Prior Institutional Research · Pre-IITMZ (2019)

Tomato volume and mass estimation using computer vision and machine learning algorithms: Cherry tomato model

Innocent Nyalala 1 , Cedric Okinda 1 , Luke Nyalala 1 , Nelson Richard Makange 1 , Qi Chao 1 , Chao Liu 1 , Khurram Yousaf 1 , Kunjie Chen 1
1College of Engineering, Nanjing Agricultural University, Nanjing, China
Journal of Food Engineering (Elsevier), Vol. 263, pp. 288–298 (Dec 2019)
Tomato volume and mass estimation using computer vision and machine learning algorithms: Cherry tomato model

Summary: Kinect depth-imaging rig positioned 0.7 m above the 0.5 m sorting conveyor capturing multi-orientation 2D and 3D descriptors (RBF-SVM R^2 = 0.9824).

Abstract

Constructs a depth-camera computer vision pipeline extracting 2D geometric and 3D depth volume descriptors trained with support vector regression and Bayesian ANN models to estimate cherry tomato mass and volume with high precision (R^2 > 0.97).

Keywords: cherry tomatovolume estimationmass estimationdepth imagingSVRJournal of Food Engineering

Key Findings & Contributions

97.06%
Mass RBF-SVM Acc
Using 2D orientation features
96.94%
Volume RBF-SVM Acc
Combined 2D + 3D depth cues
R² = 0.9824
Power-Law Test R²
M = 1.312 V^0.9551 (RMSE 15.84 g)
178+ Cites
J. Food Engineering
Vol. 263, pp. 288-298
  1. Established the physical mass-volume power relationship M = 1.312 V^0.9551 calibrated on test samples at R² = 0.9824 (RMSE = 15.84 g).
  2. Evaluated five machine learning regressors on multi-orientation 2D and 3D depth descriptors of cherry tomatoes.
  3. Demonstrated RBF-SVM leads non-destructive grading at 0.9706 (mass, 2D features) and 0.9694 (volume, all features).

Benchmark Results

Cherry Tomato Mass & Volume Estimation Performance (J. Food Eng. 2019)

Model / RelationshipInput FeaturesPerformance Metric
RBF-SVM Regressor (Mass)2D geometric image descriptors97.06% prediction accuracy
RBF-SVM Regressor (Volume)Combined 2D + 3D depth descriptors96.94% prediction accuracy
Allometric Power LawM = 1.312 · V^0.9551R² = 0.9824 (test RMSE = 15.84 g)

BibTeX

@article{nyalala2019tomato,
  title={Tomato volume and mass estimation using computer vision and machine learning algorithms: Cherry tomato model},
  author={Nyalala, Innocent and Okinda, Cedric and Nyalala, Luke and Makange, Nelson and Chao, Qi and Liu, Chao and Yousaf, Khurram and Chen, Kunjie},
  journal={Journal of Food Engineering},
  volume={263},
  pages={288--298},
  year={2019},
  doi={10.1016/j.jfoodeng.2019.07.012},
  publisher={Elsevier}
}
APA Citation

Nyalala, I., Okinda, C., Nyalala, L., Makange, N., Chao, Q., Liu, C., Yousaf, K., & Chen, K. (2019). Tomato volume and mass estimation using computer vision and machine learning algorithms: Cherry tomato model. Journal of Food Engineering, 263, 288–298. https://doi.org/10.1016/j.jfoodeng.2019.07.012