Search Filter
- Publications
- Abstract of Theses and Dissertations
- Database
- Development of Neural Network Model for Predicting Maturity and Damage...
Development of Neural Network Model for Predicting Maturity and Damage of Durian Fruit Based on its Acoustic Characteristics
Abstract:
The objective of this study was to develop a neural network model which has the capability to estimate the maturity and ripeness of durian using destructive and non- destructive methods. The physicochemical and acoustic characteristics of durian and their relationships were determined. The results revealed that the physicochemical properties (i.e., density and hardness) and acoustic characteristics (i.e., ultrasound transmissibility [Mo]) decreased with the increase of both maturity and ripeness of durian.
Two kinds of models were developed based on the input data. The first used data from non-destructive methods as inputs for predicting data from destructive methods. The second used data from destructive methods as inputs for predicting the maturity and ripeness of durian. The models’ outputs were the following durian classifications: eight (immature, fully mature, ripe, overripe, defects-immature, defects-fully mature, defects-ripe, and defects-overripe), five (immature, fully mature, ripe, overripe, and defects), four (immature, fully mature, ripe, and overripe), and two (non-defects and defects). The models were validated using various nodes and iterations: 4, 6, 8, and 10 nodes; and 1,000 and 5,000 iterations.
The validation results showed that the model with eight nodes was the best model for predicting the maturity and ripeness of durian, yielding to four durian classifications (immature, fully mature, ripe, and overripe). The accuracy of the model was 46.7-83.3 percent when using data from non-destructive methods and 60.0-94.4 percent when using data from destructive methods as inputs. On the other hand, the model separating the wholesome durian from the ones with defects was recommended to be run in six nodes and 5,000 iterations reaching an accuracy of 84.8 percent in predicting the wholesome durian.