Deep Learning-Based Multiclass Classification of Rice Seed Germination Status Using CNN Models

The quality of rice seeds is crucial to agricultural productivity and food security. Traditional methods of determining seed germination status—classifying seeds as normal, abnormal, fresh ungerminated, or dead—are time-consuming and susceptible to human error. To address result of classifying in the conventional method, this study uses deep learning to automate the classification process, using three convolutional neural networks (CNNs): MobileNetV2, ResNet50V2, and InceptionV3. This also includes a comparison analysis in identifying the most precise model for rice seed germination assessment. A dataset of 3,328 images from Philippine inbred rice variances was expanded to 16,630 samples to ensure effective model training. All models demonstrated remarkable performance. with ResNet50V2 achieved 99% validation accuracy with no overfitting, InceptionV3 with 98% validation accuracy and strong class separation, and MobileNetV2 with 97% validation accuracy with an acceptable multi-class ROC curve of 1.00. The results show that CNN-based models are capable of classifying rice seed germination status, providing a scalable and efficient alternative to manual methods.