Rice Root Disease Detection Using Deep Learning Technique

Abstract

Numerous root diseases pose a threat to rice crops, which are essential for world food security because they can stunt crop growth and reduce harvest yields. This study looks at how different deep learning models perform when it comes to picture classification and how different augmentation strategies affect their efficiency. This study examines the accuracy and loss during training and validation of three Convolutional Neural Networks (CNNs): InceptionV3, Xception, and CNN. The significance of data augmentation in improving generalizability is further shown by considering enhanced variants of these models. In addition, a CNN architecture is fine-tuned to evaluate its effect on model performance. Training accuracy was 96.88% for the baseline CNN and 98.57% for the enhanced version. With the help of augmentation, the validation accuracy was able to achieve 84.38%. The model's resistance to overfitting was demonstrated by the fact that, although fine-tuning the enhanced CNN somewhat reduced the training accuracy (92.63%), the validation accuracy remained strong (84.38%). Despite a faultless training accuracy rate of 100%, InceptionV3 struggled with generalization at 84.38%. In response, Augmented Inception V3 struck a balance between the two, improving validation accuracy to 90.62% and training accuracy to 98.41%. In terms of training accuracy, Xception was moderate at 86.32% but had a hard time with generalization at 66.75%. With the help of augmentation, Xception was able to achieve a validation accuracy of 78.12% and a training accuracy of 96.88%. By combining great accuracy with efficient generalization, Augmented InceptionV3 stood out among these models. To further improve model performance, the paper suggests continuing to explore fine-tuning tactics, augmentation techniques, and potential ensemble approaches. This in-depth study sheds light on the inner workings of deep learning picture categorization models, helping academics and industry professionals make informed decisions when building and deploying these models.