<![CDATA[This study used smart packaging made from durian seed starch with butterfly pea flower extract and applied the DenseNet121 deep learning model for real-time quality classification. Colorimetric analysis revealed a significant transition from Hue ~92° to Hue ~289° as food quality deteriorated due to pH changes. This study aimed to introduce a non-destructive vision-based classification framework utilizing the DenseNet121 deep learning architecture to demonstrate its effectiveness as a real time food freshness monitor. The dataset was divided into 70% training set, 20% validation set, and 10% test set to ensure robust model development. Performance was evaluated using accuracy, precision, recall, and F1 score. This methodology integrated physicochemical (pH) analysis with the development of a DenseNet121 architecture based Deep Learning model to classify food freshness phases. The results showed that edible films derived from durian seed starch and butterfly pea flower extract were capable of being indicators of food freshness. The dataset consisted of 160 images captured during a 12day experiment, which was expanded using stochastic data augmentation to improve model generalization. Computationally, the model achieved convergence with a training accuracy of 96% with loss 0.14 and achieved an internal testing accuracy of 0.94. Although testing on an external dataset recorded an accuracy of 0.78 due to environmental variability. This study proposes the first integration of durian seed starch based smart packaging with DenseNet121 architecture for automatic freshness classification of lempuk durian, providing a new approach for continuous food quality monitoring. These findings provide a quantitative basis for the application of applied mathematics and computer vision in sustainable food logistics.]]>
Copyrights © 2026
