Enhancing Precision Agriculture: A Hybrid Approach For Paddy Seed Classification And Fraud Detection

Abstract

Accurate identification of paddy seed varieties is essential in agriculture, not only for consumer protection against fraudulent labeling but also for supporting optimal crop performance. This paper introduces a novel approach for classifying paddy seeds, specifically distinguishing between "jasmine" and "gonen" varieties, through a hybrid technique combining image processing and machine learning. Our method uses a convolutional neural network (CNN) model trained on a dataset stored in CSV format, containing features such as Area, Major Axis Length, Minor Axis Length, Eccentricity, Convex Area, Equivalent Diameter, Extent, Perimeter, Roundness, and Aspect Ratio. For classification, an image containing multiple paddy seeds is provided as input, where each seed is segmented using preprocessing steps including grayscale conversion, morphological operations, and watershed segmentation. The CNN model then extracts the aforementioned features from each seed in the image and classifies it as "jasmine" or "gonen." Seeds that do not match the trained classes are labeled as "unknown." This robust classification tool enables both consumers and farmers to verify seed authenticity, thereby ensuring that high-quality seed varieties are planted. Experimental results demonstrate high classification accuracy, highlighting the potential of this system as a scalable and accessible tool for quality control in seed markets, benefiting both agricultural production and consumer trust.