Traditional Machine Learning and Deep Learning Modeling for Legume Species Recognition

Legumes or beans are protein-rich vegetables that are cultivated in many parts of the world. More than fifty local legume varieties are available in Bangladesh, where all these legumes do not have the same level of nutrition and taste. In this paper, classic machine learning and deep learning techniques are used to develop a machine learning model for legume species recognition. After discussing with some researchers of Bangladesh Agricultural Research Institute (BARI) about the essential factors of legume species, we start collecting data from BARI, local markets, legume fields, and the Internet. After the completion of data collection, image preprocessing, i.e. resizing and augmentation takes place. Then the images are segmented out and some features are extracted to be fed to traditional machine learning models. To categorize the legume species, three classic machine learning models, k-nearest neighbors, support vector machines, and decision trees, and three deep learning, i.e. convolutional neural network models, VGG-16, Inception-v3, and ResNet-50, are utilized. The corresponding parameters of each of these models are tuned to reach the optimum configured model. The performances of each of these models are evaluated using six noteworthy performance metrics. Inception-v3 outflanks all other models not only exhibiting an accuracy of about 98.6% but also providing the best results in terms of the rest five indicative performance metrics. © 2022, The Author(s), under exclusive licence to Springer Nature Singapore Pte Ltd.