Machine Learning based Identification of Spices: A Case Study of Chilli

Abstract

Chilli is one of the important spices and commonly adulterated for the economic gain. Manual identification is time-taking, expensive and needs experience expert which are often limited available. To deal with these issues, demand for the non-destructive methods for measuring the quality of the spices such as machine learning based automatic model were the main motives behind this study. This research paper presents the development of a deep machine learning model for the identification of chilli using a combination of convolutional neural network (CNN) and MobileNetV2 as a transfer learning model. An image dataset comprising of 4000 chilli spice images was created and made available online to create a reference database which has been further divided in 7:3 in training and validation datasets. To overcome the challenge of large datasets, transfer learning was employed during model formation. The developed model achieved an impressive accuracy of 97.97% on training dataset, demonstrating its ability to correctly classify chilli spice samples. The confusion matrix provided insights into the model's classification results, showing a higher accuracy in identifying chilli spice. Although some misclassifications occurred, the training and validation accuracy data demonstrated the model's learning progress, reaching a validation accuracy of 98.7% by the 14th epoch. The developed CNN-MobileNetV2 model showcased excellent performance, offering reliability and accuracy in chilli spice classification. Moreover, the proposed model can be extended to identify other spices with similar visual characteristics, which would have significant implications in the spice industry.