A Deep Learning Based Approach for Automated Detection of Seed Health Conditions

Abstract

Seed health plays a critical role in agricultural productivity, as diseased seeds negatively affect germination and yield. Manual inspection is time-consuming and unsuitable for large-scale evaluation. This study proposes a deep learning framework for automated classification of seed images into healthy and unhealthy categories using transfer learning with pre-trained convolutional neural networks. Data preprocessing and augmentation improve robustness against variations in lighting, rotation, and background. A structured training-testing pipeline prevents data leakage and ensures unbiased evaluation. Performance is assessed using accuracy, precision, recall, F1-score, ROC-AUC, and confusion matrices. Experimental results demonstrate reliable classification performance, highlighting the effectiveness of transfer learning for scalable seed quality assessment in precision agriculture.