Intelligent Modeling of Bread texture Acoustic Measurement Method and Artificial Neural Network (Case Study: Enriched Bread with Chia)

The purpose of this study is to model the bread texture by acoustical- mechanical method, non- destructively. To do so, bread texture enriched with 3 different levels of modified chia (2.5%, 5% and 7.5%) was evaluated by texture analyzer at a test speed of 3 mm. s-1 to 30% compression, while the microphone was located at 5 cm from the samples at a 45° angle to the horizon. From the sound recorded during loading mean sound intensity, maximum of sound, variance, standard deviation, root mean absolute value, root mean square value, skewness, kurtosis, fifth moment, sixth moment, energy, entropy in time domain, and spectral entropy and natural frequency in the frequency domain were extracted. After selecting the most suitable features (maximum sound, variance, standard deviation, root mean absolute value, energy, entropy and natural frequency), based on statistical analysis, artificial neural network with 3 algorithms ( Marquardt, scaled conjugate gradient, gradient descent) was trained and tested with 7 neurons in the input layer (in accordance with the selected features) and 3 neurons in the output layer (hardness, gumminess, chewiness). Based on the results, the training error in the Levenberg - Marquardt algorithm was lower than the other algorithms, and the root mean squared error of the test stage of this algorithm to predict hardness, gumminess and chewiness were 0.14, 0.23, and 0.33, respectively. This shows the ability of the proposed method in predicting the quality of bread.