IDLRN-DBN: SEGMENTATION-BASED EARLY DIAGNOSIS OF RICE PLANT DISEASE DETECTION USING DEEP BELIEF NETWORK

Abstract

Agriculture remains the basis of the Indian economy, with rice being a pivotal crop. However, rice cultivation faces significant challenges from various plant diseases, leading to substantial agricultural losses. The advanced segmentation-based Deep Belief Network (DBN) model iDLRN-DBN is proposed to meet the need for early and precise diagnosis of rice plant diseases and addresses the critical challenge of agricultural losses caused by plant diseases in India. The methodology is further improved upon as compared with the traditional methods by the addition of deep learning (DL), along with being supported by machine learning (ML) techniques. The six critical phases are comprised of the iDLRN-DBN model: image acquisition from a Kaggle dataset, pre-processing using adaptive filters, image segmentation through an enhanced DeepLabV3+ model, feature extraction through Efficient Grey Level Co-occurrence Matrix and Haralick Texture Features, a hybrid optimization model known as the Lyrebat Algorithm for optimal feature selection, and the classification of disease through a DBN. The iDLRN-DBN model reduced the false negative rate to 0.0257, the false positive rate to 0.0245, precision was boosted by 0.0050, and the negative predictive value increased by 0.0926. This model showed high accuracy at 0.9762, F1 score at 0.9860, low FNR at 0.0250, and a high NPV at 0.8676, which implies it is useful in conducting timely interventions and efficient crop management in smart farming. This research contributes to the progress of agricultural technology and food security due to the integration of techniques in ML and DL into the diagnostics of rice plant diseases.