Indigenous Knowledge Based Enset Crop Medicinal Value Identification Using Computer Vision

Abstract

Enset (Ensete ventricosum) is a home garden crop known for food security and has a significant traditional medicinal value in Ethiopia ,particularly in Wolaita. However, this set of knowledge is largely anecdotal and is at risk from declining oral culture and lack of scientific validation. The division between conventional understanding and scientific endorsement has restricted the wider pharmacological use of Enset. In view of this fact, a computer vision model is needed, which has been used in previous studies for the identification of the plant part usage. To achieve the objective of the study, the researchers employed an experimental research design with a qualitative and quantitative approach. Based on the Purposive sampling, important data were collected from agricultural field and research institute in Wolaita area for the development of the model. This study applied pre-trained computer vision-based models such as MobileNet, DenseNet, and NASNet to classify Enset crop parts based on their medicinal values from image data that were labeled in collaboration with indigenous communities and botanist. Based on the experiment conducted,the DenseNet model outperformed other models, with a Hamming loss of 0.001, subset accuracy of 99. 08%, and F1 scores of more than 99. 6%, making it the most suitable model. NASNet also exhibited good performance with a Subset Accuracy of 99. 19% and mean average precision of 0.999, compared to 97. 64% MobileNet Subset Accuracy. The results indicate that the computer vision model, particularly the DenseNet model, can successfully classify the medicinal value of Enset.