Search

Lemons are fruit products that grow well in Indonesia. Montaji Agrihorti is one of the lemon varieties found in Indonesia, a new variety developed by Balitjestro breeding. This lemon variety is seedless. In fact, lemons are harvested nearly all year-round. Equally important, evaluating the fruit's maturity level is crucial for determining the optimal harvest time. In this study, standardizing measurement on maturity level was conducted through Vis-NIR spectroscopy and machine learning models. In this case, non-destructive data from Vis-NIR spectroscopy were correlated with parameters related to fruit maturity and quality, such as soluble solid content (SSC), acidity, firmness, essential oil yield, and essential oil content. Non-destructive test involved capturing spectral data to be subsequently processed through machine learning models such as SVM, KNN, and random forest. The most accurate results were obtained using the SVM method for SSC and firmness parameters, with accuracy of 72 and 78%, respectively. For visual and acidity parameters, the most accurate result was performed through random forest with visual accuracy value 94% for all features, all features-MA (moving averages) was 97%, 36-PCA (principal component analysis) was 94%, and 36-PCA-MA was 97%. As for acidity, the accuracy for all features was 89%, all features-MA was 81%, 36-PCA was 89%, and 36-PCA-MA was 83%.

Lemons are fruit products that grow well in Indonesia. Montaji Agrihorti is one of the lemon varieties found in Indonesia, a new variety developed by Balitjestro breeding. This lemon variety is seedless. In fact, lemons are harvested nearly all year-round. Equally important, evaluating the fruit's maturity level is crucial for determining the optimal harvest time. In this study, standardizing measurement on maturity level was conducted through Vis-NIR spectroscopy and machine learning models. In this case, non-destructive data from Vis-NIR spectroscopy were correlated with parameters related to fruit maturity and quality, such as soluble solid content (SSC), acidity, firmness, essential oil yield, and essential oil content. Non-destructive test involved capturing spectral data to be subsequently processed through machine learning models such as SVM, KNN, and random forest. The most accurate results were obtained using the SVM method for SSC and firmness parameters, with accuracy of 72 and 78%, respectively. For visual and acidity parameters, the most accurate result was performed through random forest with visual accuracy value 94% for all features, all features-MA (moving averages) was 97%, 36-PCA (principal component analysis) was 94%, and 36-PCA-MA was 97%. As for acidity, the accuracy for all features was 89%, all features-MA was 81%, 36-PCA was 89%, and 36-PCA-MA was 83%. [ABSTRACT FROM AUTHOR]