Irani, M., Shafafi, M., Irani, H. (2015). Monitoring the osmotic dehydration process of quince by the novel fusion modular neural networks - fuzzy logic (Fmnn-Fl). , 11(3), 247-259. doi: 10.22067/ifstrj.v1394i11.42617
Mahdi Irani; Masoud Shafafi; Hasan Irani. "Monitoring the osmotic dehydration process of quince by the novel fusion modular neural networks - fuzzy logic (Fmnn-Fl)". , 11, 3, 2015, 247-259. doi: 10.22067/ifstrj.v1394i11.42617
Irani, M., Shafafi, M., Irani, H. (2015). 'Monitoring the osmotic dehydration process of quince by the novel fusion modular neural networks - fuzzy logic (Fmnn-Fl)', , 11(3), pp. 247-259. doi: 10.22067/ifstrj.v1394i11.42617
Irani, M., Shafafi, M., Irani, H. Monitoring the osmotic dehydration process of quince by the novel fusion modular neural networks - fuzzy logic (Fmnn-Fl). , 2015; 11(3): 247-259. doi: 10.22067/ifstrj.v1394i11.42617
Monitoring the osmotic dehydration process of quince by the novel fusion modular neural networks - fuzzy logic (Fmnn-Fl)
1Ferdowsi University of Mashhad International Campus
2Department of Food Science & Technology, Sabzevar Branch, Islamic Azad University, Sabzevar, Iran
3Department of Computer Engineering, Islamic Azad University, Mashhad Branch, Mashhad, Iran
Abstract
This paper presents a novel approach to monitor food process based on Modular Neural Networks (MNNs) and fuzzy inference system. The proposed MNN consists of three separate modules, each using different image features as input including: edge detection, wavelet transform, and Hough transform. The sugeno fuzzy inference system was used to combine the outputs from each of these modules to classify the images of quince during osmotic dehydration process. To test the method, for classification, database was made of 108 quince samples’ images (12 classes). In experiments, the developed architecture achieved 91.6% recognition accuracy. Next step, solid gain, water loss and moisture content of quince samples were considered as MNNs outputs, whereas osmotic dehydration time and classified images were MNNs inputs. The minimum %MRE (18.153) with 89% prediction ability for water loss (WL) was obtained when applying two hidden layers with 6 neurons per each two layers. The lowest %MRE (35.5335) with 93% prediction ability for solid gain (SG) was obtained when using 6 and 8 neurons per first and second layer, respectively. And finally %MRE was at least (7.4759) with 96% prediction ability for moisture content (MC) by 6 and 5 neurons per first and second layer, respectively. The results show that this model could be commendably implemented for quantitative modeling and monitoring of food quality changes during osmotic dehydration process.
