Abstract
The impact of deep learning (DL) is substantial across numerous domains, particularly in agriculture. Within this context, our study focuses on the classification of problematic soybean seeds. The dataset employed encompasses five distinct classes, totaling 5513 images. Our model, based on the InceptionV3 architecture, undergoes modification with the addition of five supplementary layers to enhance efficiency and performance. Techniques such as transfer learning, adaptive learning rate adjustment (to 0.001), and model checkpointing are integrated to optimize accuracy. During initial evaluation, the InceptionV3 model achieved 88.07% accuracy in training and 86.67% in validation. Subsequent implementation of model tuning strategies significantly improves performance. Augmenting the architecture with additional layers, including Average Pooling, Flatten, Dense, Dropout, and Softmax, plays a pivotal role in enhancing accuracy. Evaluation metrics, including precision, recall, and F1-score, underscore the model’s effectiveness. Precision ranges from 0.9706 to 1.0000, while recall values demonstrate a high capture rate across all classes. The F1-score, reflecting a balance between precision and recall, exhibits remarkable performance across all classes, with values ranging from 0.9851 to 1.0000. Comparative analysis with existing studies reveals competitive accuracy of 98.73% achieved by our proposed model. While variations exist in specific purposes and datasets among studies, our model showcases promising performance in soybean seed classification, contributing to advancements in agricultural technology for crop health assessment and management.
