Improved Detection and Classification of Multiple Tasks in Paddy Crops Using Optimized Deep Belief Networks with Bidirectional Long-Short Term Memory

Existing methods for detecting and classifying pests, diseases, and nutrient deficiencies in paddy crops often suffer from several significant limitations. Primarily, these methods typically focus on detecting only one of these issues at a time, rather than providing a comprehensive solution that addresses all three simultaneously. Additionally, many existing approaches have limited accuracy are computationally intensive, and lack robustness across diverse environmental conditions and varying image qualities. These deficiencies hinder their practical application in real-world agricultural settings. This manuscript presents an improved detection and classification of Pests, Diseases, and Nutrient Deficiencies in Paddy Crops using Optimized Deep Belief Networks with Bidirectional Long Short-Term Memory (C-PDND-PC-DBN-Bi-LSTM). Initially, the input paddy crop images are taken from the dataset. Then by using the proposed preprocessing technique as Anisotropic Guided Filtering (AGF), the input data is cleaned and normalized to make it suitable for classification. Hence the preprocessed images are provided to adaptive and concise empirical wavelet transform (ACE-WT) for feature extraction. Then the weight parameter of ACE-WT is optimized using a hybrid optimization technique like Slime Mould Optimization and Golden Eagle Optimization (SMO-GEO) algorithm. Then the optimized features are fed to Deep Belief Networks with Bidirectional Long Short-Term Memory (DBN-Bi-LSTM) for Paddy Crops Diseases, pests, and nutrient deficiencies Classification. The proposed method is implemented in MATLAB. Here the performance of the proposed method is assessed using performance metrics like accuracy, sensitivity, specificity, precision, recall, F1-Score, computational time, and Receiver Operating Characteristic (ROC). The proposed method provides 23.43%, 17.99%, 36.71% higher accuracy and 33.98%, 12.54%, 19.21%, and 13.44% higher Area Under the Curve (AUC) compared with existing methods like Bidirectional Long Short-Term Memory (Bi-LSTM), Deep Belief Network (DBN), Convolutional Neural Network (CNN), and Multilayer Perceptrons (MLP) respectively.