Abstract: Seed quality assessment is a fundamental step in ensuring high agricultural productivity and food security. However, traditional methods for evaluating seed quality—relying on manual inspection and mechanical processes—are time-consuming, labor-intensive, inconsistent, and often prone to human error. In this project, we introduce a deep learning-based approach that utilizes Convolutional Neural Networks (CNNs) in combination with the YOLOv5 object detection algorithm to automate and enhance the seed quality grading process.
The proposed system focuses on five commonly used food pulses: maize, rice, beans, channa, and wheat. By analyzing characteristics such as size, shape, texture, and color from high-resolution images, the model identifies and classifies seeds into three distinct categories: Grade A (good), Grade B (fair), and Grade C (poor). The implementation leverages Python, PyTorch, Flask, and OpenCV for data preprocessing, model training, interface development, and live camera-based inference.
Real-time performance is achieved using a lightweight Flask-based GUI that enables users to conduct seed analysis via webcam with instant feedback. The model demonstrates high reliability and accuracy—achieving a performance score of 92%—even under varying lighting conditions and image quality. The system is optimized to run on low-resource devices, making it deployable in field environments as well as small-scale processing units.
This intelligent solution addresses a critical need in precision agriculture by significantly reducing human effort, improving consistency, and increasing the speed and efficiency of seed sorting. It serves as a scalable, low-cost, and practical tool that can be extended to other crop varieties, contributing toward the modernization and automation of agricultural practices.
Keywords: Seed Quality, CNN, YOLOv5, Deep Learning, Image Processing, Agriculture AI
|
DOI:
10.17148/IJARCCE.2025.14663
