Quality detection of common beans flour using hyperspectral imaging technology: Potential of machine learning and deep learning

Carbohydrate content is one of the most crucial factors in common beans flour (CBF) quality after processing. However, the analysis procedure necessitates the time-consuming and costly selection of elite genotypes from many experimental lines in a destructive manner. Combining hyperspectral imaging (HSI) with machine learning (ML) algorithms provides an effective and fast approach for evaluating the quality of food products. This study determined the quality of CBF by evaluating the contents of carbohydrate using HSI technology. The samples of this work were composed of 12 varieties CBF and each variety was treated by hydration-dehydration method. After various spectral preprocessing steps, spectral features were extracted from the spectral profiles using different feature extraction methods. Partial least square regression (PLSR), Support vector machine regression (SVMR) and Temporal convolutional network-attention (TCNA) were established to predict the contents of carbohydrate in CBF. The best value of R2 and the RMSE and RPD were 0.982, 0.165 and 4.905, respectively by topology of OSC-CARS-TCNA. The outputs demonstrated although deep learning presents more accuracy than ML models, the applied ML models not only provided acceptable and reliable accuracy but also affect significantly in time-analyzing. In addition, visualization output of the current research revealed that the developed models and system can integrate to some intelligent sensors for digitalization aims. This study demonstrates the combination of HSI and ML can be an effective tool in improving the CBF processing industry and providing sustainable and efficient methods in the production of CBF.