Classification and Localization of Diverse Rice-Grain Images Utilizing a Region Proposal-Based Transfer Learning Methodology

Image-based approaches, acting as nondestructive and rapid techniques, merge image analysis and machine learning techniques to attain automatic inspection and evaluation, for discriminating and classifying varieties of grains through morphological, color-related and textural traits, or within a combination. Emerging as a significant application in agriculture, image classification manifests its significance in tasks like plant recognition, localization and classification where deep learning models of MobileNetV2 and Xception have demonstrated considerable efficacy. Accordingly, the current research investigates the efficacy of deep learning models in accurately classifying rice varieties (i.e. Osmancık97, İskender, Rekor, Yatkın and Gala, which are cultivated in Türkiye), providing significant contributions in terms of improving quality control and efficiency within the agricultural sector. By evaluating the effectiveness of MobileNetV2 and Xception models specifically for rice classification, the study generates analyses which indicate that both MobileNetV2 and Xception models achieve high levels of accuracy and sensitivity. The model proposed points toward a systematic approach to using convolutional neural networks and machine learning algorithms. Furthermore, the results emphasize the effective use of deep learning architectures in rice classification tasks in the current research which examines the performance of the deep learning (DL) models when integrated with classification methods including Support Vector Machine (SVM), eXtreme Gradient Boosting (XGBOOST), K-Nearest Neighbors (KNN) and Random Forest (RF). A thorough evaluation of the models’ performance and computational efficiency is carried out through analyzing metrics such as accuracy, precision, specificity, sensitivity as well as F1 score. Taken together, it has been demonstrated that DL models, namely MobileNetV2 and Xception are well-aligned with rice classification tasks, and their performance can be enhanced through the integration of various classification methods. These outcomes derived from the current research carried out represent a significant advancement for agricultural applications and related domains towards the development of automated plant recognition and classification systems.