Scholarlink

<p>Nematodes are microscopic organisms that exhibit dual roles in agriculture: certain species act as biological insecticides, while others cause significant crop damage, reducing global agricultural yields by 10.0–20.0% annually. Accurate identification of nematode species is critical for leveraging their beneficial properties or mitigating their harmful effects. Effective crop management depends heavily on the accurate distinction between non-parasitic and plant-parasitic nematodes. However, conventional identification techniques are often time-consuming, labor-intensive, error-prone, and dependent on sophisticated laboratory equipment. To overcome these limitations, this study proposes an automated classification system based on deep learning techniques to streamline and enhance nematode identification. The main goal of this work is to develop and assess an automated DL framework for the precise classification of plant-parasitic nematodes by designing a custom convolutional neural network (CNN) and utilizing pre-trained models—MobileNetV3, VGG16, and InceptionV3—to evaluate and compare their performance in identifying nematode species from microscopic imagery. The work is divided into two phases: (1) developing a Convolutional Neural Network (CNN) model from scratch and (2) utilizing pre-trained CNN models, including InceptionV3, VGG16, and MobileNetV3, to enhance classification performance. The approach was validated using a dataset comprising microscopic images of six plant-parasitic nematode species—<i>Acrobeles Sp.</i>,<i> Acrobeloides Sp.</i>,<i> Aphelenchoides Sp.</i>,<i> Amplimerlinius Sp.</i>,<i> Aporcelaimus Sp.</i>,<i> and Axonchium Sp.</i> taken from the I-Nema dataset. The original dataset of 1497 images was expanded to 6000 images using data augmentation techniques. The custom CNN model trained for 50 epochs using 65% of the augmented dataset while 10% used for validation. The classification performance was tested on a randomly selected subset of 250 images per class from the remaining 25% of the dataset, achieving an average accuracy of 92.06%. Subsequently, the pre-trained CNN models demonstrated superior performance, with accuracy rates of 96% (InceptionV3), 98.00% (MobileNetV3), and 90% (VGG16).</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

Deep Learning Based Automated Classification of Plant-Parasitic Nematodes Using a Proposed CNN Model and Pre-Trained Networks: MobileNetV3, VGG16, and InceptionV3

  • 期刊论文

  • 发表日期：

    2025年7月28日

查看全文
  • Meetali Verma,
  • Ayushi Kotwal,
  • Jatinder Manhas,
  • Vinod Sharma

摘要

Nematodes are microscopic organisms that exhibit dual roles in agriculture: certain species act as biological insecticides, while others cause significant crop damage, reducing global agricultural yields by 10.0–20.0% annually. Accurate identification of nematode species is critical for leveraging their beneficial properties or mitigating their harmful effects. Effective crop management depends heavily on the accurate distinction between non-parasitic and plant-parasitic nematodes. However, conventional identification techniques are often time-consuming, labor-intensive, error-prone, and dependent on sophisticated laboratory equipment. To overcome these limitations, this study proposes an automated classification system based on deep learning techniques to streamline and enhance nematode identification. The main goal of this work is to develop and assess an automated DL framework for the precise classification of plant-parasitic nematodes by designing a custom convolutional neural network (CNN) and utilizing pre-trained models—MobileNetV3, VGG16, and InceptionV3—to evaluate and compare their performance in identifying nematode species from microscopic imagery. The work is divided into two phases: (1) developing a Convolutional Neural Network (CNN) model from scratch and (2) utilizing pre-trained CNN models, including InceptionV3, VGG16, and MobileNetV3, to enhance classification performance. The approach was validated using a dataset comprising microscopic images of six plant-parasitic nematode species—Acrobeles Sp., Acrobeloides Sp., Aphelenchoides Sp., Amplimerlinius Sp., Aporcelaimus Sp., and Axonchium Sp. taken from the I-Nema dataset. The original dataset of 1497 images was expanded to 6000 images using data augmentation techniques. The custom CNN model trained for 50 epochs using 65% of the augmented dataset while 10% used for validation. The classification performance was tested on a randomly selected subset of 250 images per class from the remaining 25% of the dataset, achieving an average accuracy of 92.06%. Subsequently, the pre-trained CNN models demonstrated superior performance, with accuracy rates of 96% (InceptionV3), 98.00% (MobileNetV3), and 90% (VGG16).