Deep learning-based disease detection in peanut cultivars utilizing transfer learning

Abstract

Abstract Agriculture is a highly dynamic area that sustains global food security, and crop health plays a critical role in obtaining agricultural output. Peanuts, commonly known as groundnuts, hold a significant value due to their nutritional importance and economic benefits in many regions. However, it faces numerous challenges due to its vulnerability to a range of diseases that have a severe impact on both quality and yield. Disease detection methods based on traditional techniques were time-consuming, vulnerable to human mistakes and Labor-intensive. To handle these issues, we suggest a cutting-edge technique for the detection of peanut diseases using deep learning models. In this research work, we initially proposed some pre-trained deep learning models such as EfficientNet-B0, EfficientNet-B4, and ConvNeXt-Base, but none of them produced state-of-the-art results. To address this challenge, we leveraged the ResNet-50 Architecture using transfer learning enriched with the combination of advanced methodologies such as Data augmentation, OneCycleLR Learning rate scheduler, and weighted loss. This approach dramatically boosted the performance across various metrics, demonstrating the strength of transfer learning to handle imbalanced data and refine generalization. Deep learning models were trained with 1720 publicly available images of the dataset. The dataset includes both healthy and diseased images of groundnut leaves, including Alternaria leaf spot, Rosette, Rust and Leaf spot (early and late). The dataset was partitioned into 80% training images, 10% test images, and 10% value accuracy images. Our proposed methodology outperformed on the same dataset and gave better results than the older ones on the same dataset. The earlier same dataset had an accuracy rate of 96.51%. Our experiments showed that the suggested technique achieved a 97.28% accuracy rate and outperformed the existing state-of-the-art models. Results from these experiments demonstrate the merit of the proposed model for application in real-world agricultural problems, establishing a new baseline for detecting groundnut leaf diseases and establishing the feasibility of AI-based solutions for enhancing transferable sustainable agricultural practices.

Plant phenotyping relevance

落花生葉の病徴を画像から分類する深層学習手法を開発・評価しており、植物の病害状態推定が研究の中心であるため含める。

Code and data availability

The paper trains models on a publicly available groundnut leaf disease dataset (1,720 images, Sasmal et al.), but that dataset is cited prior work [6] rather than an authors' deposit, and no authors' code, models, or data availability URL is provided in the supplied blocks. No paper-specific public asset with an author

No evidence-backed public reproduction asset is currently recorded.