Scholarlink

<p>‘Kinnow’, the third most widely cultivated fruit in India, possesses enormous economic significance. Assessing the quality of ripe ‘Kinnow’ fruit or disease identification is crucial to ensure that quality fruits reach markets. Traditionally, quality assessment or disease identification is performed manually, which is a&#xa0;challenging, laborious and tedious task. In recent years, several research studies implementing machine and deep learning-based techniques have been introduced to automate the process of disease identification in ‘Kinnow’. However, there are several limitations associated with these recently introduced techniques, such as difficulty in segregating ‘Kinnow’ with leaves, lack of publicly available datasets (ripe fruits), limited accuracy and many more. Considering these issues, the current approach involves initially curating a&#xa0;novel dataset of high-quality ‘Kinnow’ images, collected in a&#xa0;farm setting under varying lighting conditions. Subsequently, several crucial steps including data labeling into three distinct categories, preprocessing and augmentation are applied to improve the data quality. Finally, a&#xa0;deep learning-based dual-stage fusion architecture integrating transformers with EfficientNet is proposed to accurately identify and classify the ‘Kinnow’ images into respective healthy and disease (insect attack and fungal) categories. The proposed approach is compared with state-of-the-art ‘Kinnow’ classification models to demonstrate the accuracy benefits. From the evaluation results, it is observed that the proposed approach performs better than existing benchmark techniques, achieving an accuracy of 0.925, an average precision of 0.947, an average recall of 0.927, and an average F1-score of 0.936.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

A Deep Learning-Based Dual-Path Fusion Architecture for Automated Disease Identification in ‘Kinnow’

  • 期刊论文

  • 发表日期：

    2025年11月18日

查看全文
  • Aayush Asrey,
  • Jatin Bedi,
  • Simran Setia

摘要

‘Kinnow’, the third most widely cultivated fruit in India, possesses enormous economic significance. Assessing the quality of ripe ‘Kinnow’ fruit or disease identification is crucial to ensure that quality fruits reach markets. Traditionally, quality assessment or disease identification is performed manually, which is a challenging, laborious and tedious task. In recent years, several research studies implementing machine and deep learning-based techniques have been introduced to automate the process of disease identification in ‘Kinnow’. However, there are several limitations associated with these recently introduced techniques, such as difficulty in segregating ‘Kinnow’ with leaves, lack of publicly available datasets (ripe fruits), limited accuracy and many more. Considering these issues, the current approach involves initially curating a novel dataset of high-quality ‘Kinnow’ images, collected in a farm setting under varying lighting conditions. Subsequently, several crucial steps including data labeling into three distinct categories, preprocessing and augmentation are applied to improve the data quality. Finally, a deep learning-based dual-stage fusion architecture integrating transformers with EfficientNet is proposed to accurately identify and classify the ‘Kinnow’ images into respective healthy and disease (insect attack and fungal) categories. The proposed approach is compared with state-of-the-art ‘Kinnow’ classification models to demonstrate the accuracy benefits. From the evaluation results, it is observed that the proposed approach performs better than existing benchmark techniques, achieving an accuracy of 0.925, an average precision of 0.947, an average recall of 0.927, and an average F1-score of 0.936.