Fish fillet is a corruption-sensitive product that requires several quality and hygiene reviews
before consumption. Many methods of fish quality and freshness determined based on
destructive and time-consuming tests require manpower and sample sizes. This study
evaluated the ability of a non-destructive fusion method of ultrasound, electrical resistance
measurement, and colorimetry to monitor the freshness of rainbow trout fillets at refrigerator
temperature. For this purpose, the samples of fish fillets were prepared and stored in a
refrigerator for 12 days, and physical tests such as ultrasound wave attenuation, electrical
resistance, and colorimetry to determine the color component of the fillets in both color space
of RGB and Lab, and sensory evaluation (color, taste, odor, and texture) were performed.
Also, a chemical test was conducted to determine the fillets ash, protein, and fat during
storage. Extracted data from physical tests were subjected to two classification models of
neural network and decision tree for determining the best classification method according to
the fillet freshness. Various single hidden layer neural networks with zero to ten neurons and
decision trees with five algorithms of LMT, REP, J48, RANDOM FOREST, and RANDOM
TREE were evaluated. In each model, physical test data were considered as input and storage 
days as output. Based on the results, the highest classification accuracy (100%) and the lowest
mean error value (0.02) were obtained for the neural network with 6 neurons in the hidden
layer, and correspondingly 100 percent and nearly zero for the decision tree with J48
algorithm The confusion table of both models also confirmed the classifiers' accuracy. The
integrated non-destructive system presented in this study does not require sample preparation. 
It is accurate, relatively inexpensive, and can be used during the food and fishery process to
determine their freshness.