Visual Phenotype Manifold Reconstruction for Early Foliar Abnormality Analysis in Cauliflower Imaging Systems

Abstract

Early identification of leaf-related infections in cauliflower is crucial for reducing crop damage and ensuring stable agricultural output. Traditional inspection techniques, which depend on human observation, are often inconsistent, labour intensive, and unsuitable for large farming environments. To overcome these challenges, this research introduces an intelligent hybrid framework combining Machine Learning (ML), Deep Learning (DL), and Transfer Learning (TL) for automated cauliflower leaf disease recognition. The proposed Cauliflower Leaf Disease Classification (CLDC) system utilizes a curated dataset categorized into eleven disease classes. Image preprocessing involves resizing to 64×64 pixels and normalization to enhance model performance. A novel Inception Residual Networkbased Convolutional Neural Network (IRN-CNN) is designed to extract high-level discriminative features using customized inception-residual modules. These deep features are further processed using Logistic Regression (LR) to improve classification accuracy and generalization. For performance benchmarking, conventional models such as Decision Tree Classifier (DTC), Artificial Neural Network (ANN), and standalone LR are also implemented. The system is integrated into a Tkinter-based Graphical User Interface (GUI), enabling functionalities such as dataset upload, preprocessing, training, evaluation, and real-time prediction. Batch image analysis with CSV export support enhances usability for large-scale applications. Additionally, an Explainable Artificial Intelligence (XAI) component powered by a generative AI API provides detailed insights, including disease severity, affected regions, and crop verification. A Telegram Bot interface further extends accessibility for mobile-based detection. Experimental findings confirm that the proposed IRN-CNN hybrid model delivers superior accuracy and reliability, making it a scalable solution for smart agriculture and precision farming systems.

Plant phenotyping relevance

カリフラワー葉の病害状態を画像から分類・推定する手法の開発と性能比較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 describes a curated cauliflower leaf disease image dataset and an IRN-CNN classification pipeline, but no block contains any public deposit, repository, or availability statement for the dataset, code, or trained models. The only URLs present are cited prior-work DOIs (Foods 2021, Sustainability 2022), which,

No evidence-backed public reproduction asset is currently recorded.