| Abstract: | The key step in making beer is choosing the malt-barley. Every malt house requires that the types of grain be checked before being purchased. The creation of premium malt depends on varietal consistency. It might be challenging to distinguish between different malt-barley kinds during inspections because it takes knowledge and practise. To identify several kinds of Ethiopian malt-barley, a computerised image processing technique based on combined morphological, texture, and colour aspects has been investigated. The one local location for taking the varieties is the Gondar Malt Factory, where sample malt-barleys were taken. Each of the four variations yielded an average of 52 photos (Holker, Propino, Sabini and Misikal). A total of 208 pictures were captured, and each one included 5408 malt-barley seeds. Each image of a scanned barley seed was used to extract nine morphological, five texture, and six colour features for identification. K-Nearest Neighbors (KNN), Artificial Neural Networks (ANN), and the combination of the two techniques are examined for the construction of identification models for the prediction of maltbarley types. According to experimental findings, an ensemble model combining ANN and KNN performs better on combined characteristics of morphology, texture, and colour when utilising the Sequential Forward Feature Selection (SFFS) technique than a single model built using either ANN or KNN. For the Holker, Propino, Sabini, and Misikal varieties, a quantitative accuracy of 86% is attained utilising the ensemble of ANN and KNN with the combined feature sets of morphology, colour, and texture. This indicates a positive outcome for creating a useful malt-barley identification model. Malt-barley photos with non-uniform size and overlap significantly impair the effectiveness of the identifier; as a result, this area of future research requires an examination of general segmentation and noise removal techniques. |
