This work determine the volume of hams particularly for the horizontal viewpoint. Three mathematical approaches had been proposed to detect the defective images in the image sequences:
- Based on the k-nearest neighbor
- Based on minor axis
- Based on Y-direction
The method had been evaluated on 2 databases: HamDB-A and HamDB-B, where the data elicitation method are slightly different. Note that, a Mask Region–based convolutional neural network (Mask R-CNN) object segmentation approach was adopted to extract the volume of the ham object from a video.
Software is written and tested using Matlab 2020a, toolbox required:
Statistics and Machine Learning Toolbox
The files include:
- defectDetection_knn.m - defect detection based on the knn
- defectDetection_minor.m - defect detection based on the minor axis
- defectDetection_y_axis.m - defect detection based on the y direction
- curveFitting_polynomial.m - optimal curve fitting functions based on polynomial order 2
- curveFitting_power.m - optimal curve fitting functions based on power function
- input videos - Please download from https://drive.google.com/drive/folders/19A-p_KiBM15DNeJzNMiyfXiIixW3za7Q?usp=sharing
Acquisition setup for the elicitation and video recording of the ham sample:
Example of 16 ham samples in HAMDB-A:
Example of 16 ham samples in HAMDB-B:
(a) Good Mask R-CNN image and (b) Poor Mask R-CNN image:
Volume prediction result on HamDB-A (left) and HamDB-B (right):
If you use this method in your research, please cite:
@article{liong2020ham,
title={A Statistical Approach in Enhancing the Volume Prediction of Ellipsoidal Ham},
author={Y. S. Gan and Lan Wei and Yiming Han and Chenyu Zhang and Yen-Chang Huang and Sze-Teng Liong},
journal={Journal of Food Engineering},
volume={1},
pages={1--1},
year={2020},
publisher={Elsevier}
}
If you have suggestions or questions regarding this method, please reach out to stliong@fcu.edu.tw
Thank you for your interest and support.