Plant Leaf Disease Detection Using Metaheuristic Optimization Algorithms and Deep Learning

Abstract

Plant diseases significantly reduce the yield and the production of crops across the globe.Crop productivity, plant development and human access to food have all been hampered by the prevalence of plant diseases throughout the history.In general, leaves exhibit symptoms if the plant is affected by diseases.Therefore, it is essential to identify the type of infestation to reduce the destructiveness of the disease.This scenario allows one to replicate the spread of infectious diseases and the inability of farmers to recognize and remember them.One possible approach to tackle this issue is to utilise Deep Learning (DL) techniques in conjunction with Machine Learning (ML) approaches within the domain of Computer Vision (CV).The current research has introduced the APLDD-ESOSDL approach, which utilises deep learning to optimise the search for symbiotic organisms in order to automate the detection of plant leaf diseases.The objective of the proposed APLDD-ESOSDL approach is to enhance agricultural yields and reduce crop losses by offering farmers a visual depiction of disease symptoms.The goal of the APLDD-ESOSDL approach is to accurately classify the presence of leaf diseases.The APLDD-ESOSDL technique utilises the inception ResNet-v2 model as a feature extractor and the Stacked Long Short-Term Memory (SLSTM) model for classification.In addition, the hyperparameters of the SLSTM algorithm are adjusted using the Enhanced Symbiotic Organism Search (ESOS) approach.A comprehensive experiment was carried out utilising the reference data set to verify the effectiveness of the APLDD-ESOSDL approach.The APLDD-ESOSDL algorithm outperformed more advanced systems, achieving a maximum accuracy of 99.22%, precision of 98.52%, sensitivity of 98.06%, and specificity of 99.54% in experimental experiments employing six distinct cutting-edge approaches.

Plant phenotyping relevance

葉画像から植物病害を自動分類する深層学習手法を提案し、データセットで性能検証しており、植物表現型取得・判定手法が中心である。

Code and data availability

The paper's plant leaf disease detection experiments use a public Kaggle image dataset (PlantVillage, emmarex/plantdisease) with 3,503 corn leaf images across four classes, which is the direct input to the paper's phenotyping/classification measurements. No author analysis code, trained models, or supplementary assets,