Hybrid Model for Plant Disease Diagnosis: Transfer Learning with Dynamic Feature Selection

Abstract

Plant disease classification is critical in sustainable agriculture, essential for ensuring food security and reducing environmental harm. Traditional disease detection methods are often inefficient, time-intensive, and costly, highlighting the need for advanced computational solutions. This work presents a hybrid intelligent model that integrates a framework of transfer learning with bio-inspired optimization in the identification of early and accurate plant diseases. The pre-trained VGG-19 is re-trained parietally for the feature extraction process by applying the transfer learning concept. The Grey Wolf Optimization (GWO) is utilized to select relevant features from the extracted features. This work assesses and evaluates the performance of the proposed model using the Plant Village dataset (39 classes) of 14 plants ( Apple, Blueberry, Cherry, Corn, Grape, Orange, Peach, Pepper, Potato, Raspberry, Soybean, Squash, Strawberry, Tomato).From the experiment result, it is evident that the proposed model has an accuracy of 99.44%, which is higher compared to the traditional machine learning algorithm and state-of-the-art approaches. This enhancement has further shown the potential of the model in evolving how plant diseases are managed due to timely and precise interventions.