| Abstrakt: |
Researchers are investigating deep learning techniques with their automatic feature learning capabilities for automated rice disease recognition from images. The current study has developed an ensamble model exploring hybrid features i.e., hand crafted and deep features. The proposed approach utilizes a laboratory image dataset comprising 2370 rice leaf images sourced from PlantVillage, augmented with another rice disease dataset featuring the same classes for training. These classes include BrownSpot, Leaf Blast, Hispa, and Healthy images in the dataset. The model achieves an accuracy of 97.9% through k-fold cross-validation. Considering the domain shift concept, we have tested the model’s accuracy on our real field rice leaf images containing BrownSpot, Leaf Blast, and Healthy leaves. The model achieves an accuracy of 93.7% on our dataset. To give the access of automatically identifying rice disease to the village farmers having poor internet connectivity, the current work introduces “easy to use” mobile application, RiceDiseaseRec. This research paves the way for automated rice disease recognition which leads to improving food security and mitigating crop yield losses. |
Full text from SpringerLink