TWO-BRANCH DEEP CONVOLUTIONAL NEURAL NETWORKS FOR EFFICIENT TOMATO PLANT DISEASE CLASSIFICATION

Abstract

In this research work, a novel Two-Branch DCNN framework has been proposed for the classification of disease in tomato plant that efficiently works in CIE Lab colour space. By considering achromatic (L channel) and chromatic (AB channels) information in separate dedicated branches, the model is both more accurate and efficient. Experiments on Plant Village dataset have shown consistent and competitive performance, and the 20%L + 80%AB model yielded 99.48% classification accuracy, which outperformed some competitive state-of-the-art GoogLeNet (98.37%) and AlexNet (97.82%). The architecture also retains strong performance on more difficult Cropped-PlantDoc dataset, with 50%L + 50%AB achieving an accuracy of 76.91%, outperforming baselines by 6 percentage points at least, while decreasing the number of trainable parameters and floating point operations by 30-50%. An ablation study indicates that the colour and the greyscale branches influence significantly on the overall performance, and the diverse optimal filter partitioning between datasets supports that the design of the architecture is able to adapt to various imaging conditions. We demonstrate a resource-efficient application of tomato plant disease diagnosis which could be installed and used in resource-limited-solution for field-level indoor disease detection on farmers’ side with real-time analysis. An Efficient Method of Tomato Diseases Based on Deep Learning in CIE lab colour field.

Plant phenotyping relevance

トマト葉の画像から病害状態を推定する深層学習手法を開発し、複数データセットで精度比較・アブレーション評価しており、植物フェノタイピング手法が中心である。

Code and data availability

The paper uses public PlantVillage and Cropped-PlantDoc datasets, but these are generic third-party benchmark datasets, not paper-specific assets. No author code, trained models, or data availability statements appear in the supplied blocks.

No evidence-backed public reproduction asset is currently recorded.