Machine Vision Approach for Identification of Four Variant Pakistani Rice Using Multi-Features Dataset

Crops are the most important and beneficial food source in Pakistan. The demand for food has been an increase in Pakistan due to population growth. Pakistan produced 7,410 million tons of rice according to the financial year survey 2020 (FYS-2020). Pakistani rice has been cultivated in 3,304 hectares of the agricultural land zone, and it is also export around the world. Rice is also increased by 0.6% Gross Domestic Product (GDP) of Pakistan (FYS-2020). The old and manual process of rice classification is more expensive and time-consuming. In this study, we describe a machine vision approach for rice identification. We use four different varieties of rice for the experimental process such as Pakei_Kaynat, Kaynat_Kauchei, and Kauchei_Super_Banaspati and Tootaa_Kauchei (P1, P2, P3, and P4). The 100 images dataset have been used for practical work and total calculated of 400 (4 x 100) image of rice. The different process has been deploying on available datasets such as introduction, preprocessing methodology, and result discussion. A quality enhancement technique has been implementing for clarifying between rice color and shape sampling, and it is also converted color image in gray scale level. Every image has been employing six different non-overlapping regions of interest (ROI’s) and calculated a total of 2400 (6 x 400) ROI’s. Binary (B), Histogram (H) and Texture (T) features have been implemented and extract 43 features on each ROI’s and total calculated 103,200 (2400 x 43) machine learning (ML) features. Best First Search (BFS) Algorithm was used for feature optimization. Different ML classifiers are implementing for experimental process namely; Function Multi-Layer-Perception, Function SMO, Random Tree, J48 Tree, Meta Classifier via Regression and Meta Bagging. The Function Multi-Layer-Perception overall accuracy (OA) has describe better accuracy result is 99.8333%.