Precision Agriculture through Hybrid Deep Learning Ensembles and Advanced Optimization for Multiclass Paddy Disease Detection

Rice, which is extensively utilized as a staple food across the globe, assumes 
significant importance in the quest for superior rice products. However, the incidence of rice 
illnesses may prevent rice-based goods from being produced at their best and of the highest 
quality. Precision agriculture investigates the development of automated disease detection 
and categorization systems in great detail. Because rice is grown in vast, moist regions, it 
might be difficult to detect these infections. A Deep Learning-based multiclass paddy disease 
finding model (DL-MPDP) is presented in this research, for accurate identification and 
categorization of afflicted areas in paddy plants. IoT cameras are utilized to acquire 
unprocessed images of paddy fields. The images undergo preprocessing techniques such as 
LANCZOS, CLAHE, and Wavelet to increase their quality before additional analysis. Then, 
the pre-processed image can be subjected to Feature Extraction (Texture Feature, Shape 
Feature, and Edge Feature) via Onehot Encoding Technique. Then, the dimensions of the 
extracted image will be reduced via a new Kernel-based Principal Component Analysis 
(KPCA). Subsequently, from the dimensionally reduced data, the optimal features will be 
provided via the newly modified Flower Pollination Algorithm (mFPA). The Recurrent 
Neural Network with Long Short-Term Memory (HRNNLSTM) model that makes the final 
detection (presence or absence of disease) is trained using the selected optimal features. 
Moreover, to enhance the detection accuracy, the weight function of HRNNLSTM is fine
tuned using the Crayfish Optimization Algorithm.