A Deep Learning-based Fine-tuned ResNet50 Model for Soybean Seeds Multiclass Classification

A crucial aspect of modern agriculture and food security is the classification and analysis of agricultural products, particularly seeds. Soybean is a significant leguminous crop known for both its economic and nutritional worth. With the increase in soybean production, it becomes challenging to classify the seeds with the diseases. The ResNet-50 architecture is a well- known version of Convolutional Neural Networks (CNNs) which is used in this study to offer a thorough investigation of the classification of healthy soybean seeds from the diseased seeds and naturally broken seeds. The ResNet-50 model is trained at different numbers of epochs to categorize soybean seeds using a dataset collected from Kaggle. The training and evaluation procedures were methodically carried out, with frequent checks on the model’s accuracy and loss. The study includes two significant epochs, each of which reveals different performance outcomes. The model’s training accuracy at epoch 25 was 97.85%, with a corresponding loss of 0.0825. Additionally, it was also determined that the validation accuracy and loss were at epoch 25 was 86.84% and 0.3719, respectively. The training accuracy at epoch 50 has been identified as 99.89% with a loss of 0.0154 whereas validation loss of 0.3619 was recorded along with an increase in validation accuracy to 89.56% in comparison to epoch 25. These results demonstrate that the ResNet-50 model is highly efficient for multiclass classification of soybean seeds which can be used for real time applications in agriculture sector.