An integrated color and texture feature based mango ripening stage classification from thermal imaging

Mango is a prominent fruit cultivated extensively in tropical and subtropical regions, serving as a significant economic asset for many countries. As Ensuring quality for export markets is essential, it traditionally achieved through labor-intensive and time-consuming manual sorting. Recently, computer vision (CV) based sorting has gained attention due to its efficiency and accuracy. While CV models leveraging RGB images can identify raw, ripened, and defective mangoes, certain internal defects and instances where raw and ripened mangoes exhibit similar coloration pose challenges, often leading to misclassification. In this study, we explore the use of thermal imaging to classify mangoes into raw, ripened, and defective categories. Our approach combines color and texture features from thermal images, excluding temperature data, while improving classification accuracy compared to previous studies. We conducted a comparative analysis of classification performance using various traditional classifiers, including Random Forest (RF), K-Nearest Neighbors (KNN), Support Vector Machine (SVM), and Linear Discriminant Analysis (LDA). Our findings indicate that the combination of GLCM (contrast, correlation) and color features achieved a $87 \%$ accuracy rate with the SVM classifier for thermal images. It is also verified that the use of a simple Convolutional Neural Network (CNN) model has increased the accuracy up to $97 \%$. However, this improvement in accuracy can only be robustly confirmed with a larger dataset.