<![CDATA[This study aims to compare the performance of two deep learning architectures, namely MobileNetV2 and ResNet50, in classifying soybean leaf diseases based on digital images. The classification task involves four classes: Bacterial Pustule, Healthy, Sudden Death Syndrome, and Yellow Mosaic. Model performance was evaluated using a confusion matrix and classification report, including accuracy, precision, recall, and f1-score metrics. The experimental results indicate that both architectures demonstrate excellent classification performance. MobileNetV2 achieved an accuracy of 95.45%, with very few misclassifications, namely one error in the Bacterial Pustule class and one in the Yellow Mosaic class. The macro average and weighted average values of 0.9545 indicate that MobileNetV2 provides stable and balanced performance, making it an effective and efficient model, particularly for systems with limited computational resources. On the other hand, ResNet50 demonstrated superior performance by achieving perfect classification results. All test data were correctly classified, as reflected by a confusion matrix entirely along the main diagonal and precision, recall, f1-score, and accuracy values of 1.000 (100%). This result indicates that ResNet50 has very strong feature extraction and generalization capabilities in recognizing each disease class. In conclusion, ResNet50 outperforms MobileNetV2 in terms of classification accuracy and prediction precision. However, MobileNetV2 remains a highly viable alternative due to its lightweight architecture and computational efficiency while still delivering near-equivalent performance. Therefore, the selection of the model can be adapted to system requirements, where ResNet50 is suitable for maximum accuracy, while MobileNetV2 is more appropriate for deployment on resource-constrained devices.]]>
