|M.Sc Student||Elkabetz Meital|
|Subject||Fusion of Non-Destructive Methods for Quality Assessment of|
|Department||Department of Civil and Environmental Engineering||Supervisors||Professor Emeritus Itzhak Shmulevich|
|Professor Raphael Linker|
|Full Thesis text - in Hebrew|
Fruit quality is very difficult to define due to the diversity of the tissues involved and the numerous attributes required to describe quality properties.
It is generally accepted that the property which reflects best the quality of fruits texture is the so-called firmness. Fruits firmness is most commonly determined through destructive methods such as: Magness-Taylor test and the standard method (ASABE Standards S368.4) - the Parallel-Plates method. In the present research this latter method is used as a reference measurement of apple firmness.
Numerous researchers have evaluated fruit firmness using low-mass impact techniques, acoustic techniques and optical techniques.
The current research motivation is to examine fusion (combination) of these non-destructive methods as a way to improve the prediction accuracy.
Four hundred and sixty apples of four types were used at the present research in a shelf-life simulation. Two destructive methods (Magness-Taylor and Parallel-plates) and four non-destructive methods (impact, acoustic, spectrometer and hyper-spectral camera) were used.
As expected, the results of the Magness-Taylor test have shown a low correlation to shelf life while the Parallel-Plates method has shown a high correlation to shelf life.
With the low-mass impact method, a correlation of R2=0.81 for all types of apples was achieved. With the acoustic method, the results was R2=0.75. Partial least squares (PLS) analysis of the 350-1000nm spectra recorded by the spectrometer led to a correlation of R2=0.81. The results obtained using the light scattering at different wavelengths were much lower, with correlation of R2=0.51.
Fusion models were developed through multi-linear regression. Six combined models were investigated, namely acoustic & camera, acoustic & impact, acoustic & spectrometer, impact & camera, impact & spectrometer and triple fusion of impact & acoustic & spectrometer. All the models based on more than one technique resulted in a correlation coefficient equal or higher than the one obtained either technique by itself. The best dual fusion was the impact & spectrometer fusion, which gave a correlation of R2=0.89 for all types of apples. Combining impact, acoustic and spectral led to a correlation of R2=0.91 for all types of apples. Therefore, it is recommended to develop a device which measures the firmness of fruits by those three none-destructive methods simultaneously and combines the data through multi-linear regression in order to sort the fruits by their firmness on the sorting line.
The major contribution of the current work is in quantization of the advantage of data fusion from different sensors for apple texture evaluation.