Deep Learning-Based Models for Paddy Disease Classification and Segmentation: An Experimental Review

Early-stage diagnosis of paddy diseases, based on initial visible symptoms, is crucial for minimizing chemical use and preventing the spread of disease. Automated paddy disease diagnosis can facilitate this and contribute to improved crop production. Accurate segmentation of diseased regions is essential for precise severity analysis, and segmentation models can be used to address the challenges related to the complex disease patterns. This study analyses the performance and applicability of seven classification models (DenseNet, InceptionV3, Xception, MobileNet, Vision Transformer (ViT), parallel-ViT, and a custom ensemble model of DenseNet121 and Xception) and eight segmentation models (DeepLabv3+, UNet, VGG16 UNet, Inception-ResNetV2 UNet, TransUNet, Attention UNet, MultiResUNet, and Deep Residual UNet) in this domain. A comparative analysis of these models, highlighting their structural differences, advantages, and limitations, is also provided. Given the high computational demands and high data dependency of ViTs, we evaluated whether traditional models, their ensembles, or memory-efficient models could match ViT performance, and found this to be feasible. Additionally, we explored the effect of augmentation intensity on the classification models. We also identified significant research gaps within this domain. To address the lack of open-access paddy disease segmentation datasets, we created a novel dataset using image-processing techniques. The applicability of the above-mentioned segmentation models was evaluated for paddy diseases using this newly created dataset and we found that the Deep Residual UNet is the most suitable model to be used in resource-constraint applications, considering its quantitative and qualitative performance, model size, and structural advantages and limitations. Furthermore, we investigated the impact of training these segmentation models with a dataset containing more than double the number of augmented images compared to the original dataset. We observed that while the quantitative performance increased with increased data, the qualitative performance degraded in a few cases. Overall, our results can assist researchers in developing effective autonomous systems for paddy disease management.