Image Based Papaya (Carica Papaya Linn.) seed germination evaluation by ResNet50

Abstract

The researchers developed a papaya seed germination evaluation system (PSGES) using ResNet50, a convolutional neural network, to evaluate papaya seed germination potential from single seed images. Using a comprehensive dataset of 12,600 papaya seed images, they allocated 11,600 images for training (with an 80/20 training-testing split) and 1,000 images for validation. The system achieved impressive performance metrics, with an overall accuracy of 99.58% and an average processing time of 1.4705 seconds per image. The training dataset demonstrated exceptional performance with 0.9958 accuracy, 0.9980 precision, 0.9972 recall, and 0.9976 F1-score. When compared to existing seed evaluation methods in the literature, PSGES showed superior precision at 99.59%, significantly outperforming Rice (ANN) at 92.80%, Beet (NIR) at 89.00%, and Chili (ANN) at 71.71%. The study revealed a papaya seed germination rate of 84.92%, calculated from (10,000 + 20 + 677 + 3) ÷ 12,600 × 100. Notably, ResNet50 demonstrated superior performance compared to six other CNN architectures tested, including AlexNet, GoogLeNet, Inceptionv3, ResNet18, ResNet101, and VGG16, in both training and validation performance metrics.