Optimization of Mango Plant Leaf Disease Classification Using Concatenation Method of MobileNetV2 and DenseNet201 CNN Architectures

Abstract

Purpose: Mango production can be severely impacted by diseases affecting mango plants. By leveraging artificial intelligence, the agricultural sector can automate the analysis of mango leaves to monitor plant health. The goal of this research is to improve the early detection of diseases in mango leaves to allow early treatment to minimize damage to the crops. Methods: This study employs an approach of combining two pre-trained CNN architectures, namely MobileNetV2 and DenseNet201 through concatenation method. To enhance the model’s generalization ability, various image augmentation techniques were applied during the training phase. Result: The model developed in this study achieved great performance in classifying mango leaf diseases with a testing accuracy of 99.25%. This result indicates the effectiveness of the concatenation method by outperforming the accuracy of either MobileNetV2 or DenseNet201 when implemented separately. Novelty: This research introduces a novel strategy by concatenating two pre-trained CNN architectures for mango leaf disease classification, a method not previously explored in this context. The model developed from this study has the potential to serve as a tool for the early detection and treatment of mango leaf diseases.

Plant phenotyping relevance

マンゴー葉の病害状態を画像から分類するCNN手法の開発・性能評価が研究の中心であり、植物病害フェノタイピングに該当する。

Code and data availability

The paper uses the public Kaggle 'Mango Leaf Disease Dataset' (MangoLeafBD) as its image input, but no authors' public URL for the dataset, code, or trained model is provided in the supplied blocks, and no allowed URL hosts such an asset. The only URLs in the text are the journal index and a cited arXiv reference, so a

No evidence-backed public reproduction asset is currently recorded.