Prediction of fruit ripening by Artificial Neural Network based on relationship between pectin and image analysis / Aisyah Sakina Shahrin ... [et al.]
This research was focuses on the prediction of fruit ripening using artificial neural network. The main purposes of this study are to correlate pectin activity (data) with image analysis (image) of figs and to investigate the compatibility of Artificial Neural Network (ANN) in speculating the figs r...
Saved in:
| Main Authors: | , , , , |
|---|---|
| Format: | Conference or Workshop Item |
| Language: | English |
| Published: |
2020
|
| Subjects: | |
| Online Access: | https://ir.uitm.edu.my/id/eprint/81122/1/81122.pdf https://ir.uitm.edu.my/id/eprint/81122/ |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | This research was focuses on the prediction of fruit ripening using artificial neural network. The main purposes of this study are to correlate pectin activity (data) with image analysis (image) of figs and to investigate the compatibility of Artificial Neural Network (ANN) in speculating the figs ripening behaviors (stage). Ripening stages is the stage where the fruit are ready to be harvest. During this phase, every fruit will undergo the weakening of parenchyma cell wall and dissolution of middle lamella. As the result, the figs is sweetening as its reach the final stage of its development which is ripening phase. In order to analyze the changes happened between the figs, the laboratory experiment such as extraction yield (EY), brix of sugar and degree of esterification (DE) were come in handy. Those data represent the statistical input of pectin structure. Later, the information being correlated with the figs resemblance. Those method is quantitative-typed method where it is said to have numerous limitation which would affect the accuracy of the results obtained. The limitations would be time-consuming, expensive and lack of consistency as the volume of chemical and procedure of sampling are changeable since human error are commonly to happen from time to time. Thus, the solution to those limitations is Artificial Neural Network (ANN). The models used is MLP model with back-propagation algorithms with the help of learning function of Bayesian regularization and the transfer function is tangent hyperbolic. It is found that neuron number eight is the most accurate than the others neuron number since it has a high R value which is 0.97194 and low value of MSE, RMSE, MAE and MAPE which are 9.18E-13, 9.58123E-07, 3.04E-04 and 0.03% respectively. |
|---|