Enhancing the Accuracy in Classifying Rice Seed Varieties using Novel Random Forest Algorithm Compared with Support Vector Machine

In agriculture, categorizing rice seed varieties presents a formidable challenge, particularly when considering their implications for production and marketing. Rice, a fundamental dietary staple, holds substantial importance in sustaining human food resources. A multitude of rice seed varieties have been grown in various regions across the world. Accurately classifying those rice seed varieties is a formidable task due to the striking visual similarities in color, shape, and texture among seeds of different varieties. This paper employs machine vision and machine learning techniques to create a swift method for detecting and categorizing rice seeds according to their variety and purity levels. The objective of our research is to perform classification of rice seed varieties in a fast and efficient manner that can be used by farmers for cultivation. A vision-based computer system was meticulously crafted to identify four distinct rice seed varieties, namely Andhra Ponni, Ko50, IR20, and Atchaya Ponni. This system harnessed the power of 23 distinct features, comprising 12 morphological features, 8 textural features, and 3 color features, for accurate classification and analysis. Then the complete dataset is created using the calculations of the above stated features obtained from the images. The created dataset is tested with classification algorithms such as Novel Random Forest and Support Vector Machine, out of which the Random Forest algorithm gives 95.12% of accuracy. The classification results effectively meet the requirements and expectations of both producers and consumers.