A High-Performance Deep Learning Approach for Real-Time Detection of Rice Leaf Diseases Using CNN Model

The present research evaluates and compares the performance of four convolutional neural network models, VGG19, VGG16, Inception V3, and ResNet50, in classifying diseases affecting rice plants, such as brown spot, rice blast, bacterial blight, and healthy. After training and testing each model with a comprehensive dataset, the researcher analyses the performance metrics, which include precision, recall, F1 Score, and confusion matrix. The results show that VGG19 consistently outperforms other models in all evaluated metrics. In terms of precision, VGG19 also demonstrates higher results in disease classification, in particular, correctly identifying brown spot, rice blasts, and bacterial blight and misclassifying fewer plants as healthy. To be more precise, those figures for VGG19 equal 96%, 97%, and 96.5%, respectively, illustrating the high capacity of this model for disease detection. At the same time, VGG16, Inception V3, and ResNet50 deliver the same good results, although these models consistently demonstrate lower precision and recall rates. The confusion matrix proves that VGG19 reliably identifies disease and requires fewer data points to mispredict the plant as healthy. Using a unique and extensive dataset, the author proves an essential role in selecting an appropriate model for specific tasks. In the context of agriculture, a model should efficiently address disease detection tasks. At the same time, VGG19 also demonstrates high rates of false prediction in classifying healthy plants as infected, which has to be improved using other models.