Identification Of Injera Mixture Using Computer Vision And Machine Learning Approach

Abstract

Injera is a culturally significant food in Ethiopia, and the majority of the population consumes it daily. It is usually made from teff flour and different variations can include barley, corn, rice, sorghum, wheat, or a combination of these flours. However, the adulteration of Injera with harmful substances poses significant problems. When bad ingredients are mixed with teff flour or other flour, it can lead to health issues for consumers, loss of cultural identity as the traditional preparation and authenticity of Injera, and it creates challenges in marketing and promoting genuine Injera, as consumers may become wary of purchasing products that are not guaranteed to be pure and safe. Addressing these problems is crucial to ensure the preservation of cultural heritage, protect public health, and maintain the integrity of the Injera market. Identification of Injera is difficult using the naked eye due to their similar features. In recent years, machine learning and deep learning algorithms have demonstrated impressive potential in image identification. This paper proposes a hybrid approach based on the best feature extraction algorithm to classify injera mixtures. Using traditional fermentation techniques, we prepared datasets consisting of Injera samples with various combinational ratios including 10:90 and 20:80 ratios. We captured hot Injera before 1 hour and cold Injera after 24 hours. In this study, we have used Grey Level Co-occurrence Matrix (GLCM), Convolutional Neural Network (CNN), and a combination of GLCM and CNN as a feature extraction technique. Also, we have used a Support Vector Machine (SVM) and Random Forest (RF) as a classifier to design the Injera mixture identification system. We have examined different combination ratios of hot and cold (after 24 hours) frontside and backside Injera. From the experimental results, we have registered an accuracy of a combinational ratio of 10:90 frontside hot Injera, 10:90 backside hot Injera, 10:90 frontside cold Injera, 10:90 backside cold Injera, 20:80 frontside hot Injera, 20:80 backside hot Injera, 20:80 frontside cold Injera, 20:80 backside cold Injera is 87%, 86%, 93%, 92%, 91%, 95%, 98%, and 98% for SVM and 88%, 87%, 91%, 91%, 93%, 94%, 98%, and 98% for RF respectively on combined features.