Efficient quality classification of pretreated and dehydrated carrots using feature selection and intelligent models

Discriminating dehydrated carrots is crucial for quality control, energy efficiency, and consumer confidence. Artificial intelligence techniques have revolutionized quality control processes by enabling them to learn, generalize, and rapidly automate complex data structures.

This study investigated the classification of carrot samples subjected to various dehydration applications and pretreatments using visible near-infrared (Vis-NIR) reflectance and artificial intelligence techniques. Carrot slices were pretreated using ultrasound, sucrose, gum arabic, and microwave methods, followed by dehydration under different microwave power outputs (100, 200, and 300 W) and vacuum pressures (atmospheric, 200, and 400 mmHg). Vis-NIR reflectance data (325-1075 nm) were obtained using a portable spectroradiometer. Recursive feature elimination with extreme gradient boosting was implemented to identify the 50 essential wavelengths, minimizing dimensionality and processing demands. The dataset was partitioned into training, testing, and validation subsets for model development and evaluation. Nineteen artificial intelligence models were utilized to differentiate dehydrated carrot slices. The results show that the multilayer perceptron and support vector machine models achieve classification test accuracies of 97.78% and 95.56%, respectively. In contrast, the test model times for these models were 0.0156 and 0.0049 s, respectively. The convolutional neural network (CNN) model proved to be the most successful deep learning model, achieving a test accuracy of 80.04% and a test time of 0.3372 s.

Consistent feature selection and artificial intelligence approaches successfully enabled non-destructive classification of pretreated and dehydrated products. © 2025 Society of Chemical Industry.