Scholarlink

Machine Learning (ML) and Artificial Intelligence (AI) have been proven as very promising areas of the topic called image processing, especially when it comes to recognizing and categorizing different types of soils. This study will use deep learning systems, such as TensorFlow and Keras, to interpret and properly categorize images of soils. This data is composed of 902 images depicting four classes of soil: alluvial, alluvial, black, clay, and red. So, these pictures were separated into training, validation, and testing Qs. An architecture of a Deep Convolutional Neural Network (DCNN) based model was trained with and without using image augmentation techniques. Data augmentation was used to extend the dataset to increase model performance. Moreover, transfer learning through the VGG-16 model was deployed in classifying soils. Without augmenting its data, the VGG-16 model attained an accuracy of 91.7%. After training with augmented data, the model attained 97.9%. In evaluating the performance and accuracy of the classification of the deep transfer learning model, a confusion matrix was employed.

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 Soil Classification Using Image Augmentation and Transfer Learning

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Janmejay Pant,
  • Rakesh Kumar Sharma,
  • Malika Kulyal,
  • Himanshu Pant,
  • Harishchander Anandaram,
  • Kapil Joshi

摘要

Machine Learning (ML) and Artificial Intelligence (AI) have been proven as very promising areas of the topic called image processing, especially when it comes to recognizing and categorizing different types of soils. This study will use deep learning systems, such as TensorFlow and Keras, to interpret and properly categorize images of soils. This data is composed of 902 images depicting four classes of soil: alluvial, alluvial, black, clay, and red. So, these pictures were separated into training, validation, and testing Qs. An architecture of a Deep Convolutional Neural Network (DCNN) based model was trained with and without using image augmentation techniques. Data augmentation was used to extend the dataset to increase model performance. Moreover, transfer learning through the VGG-16 model was deployed in classifying soils. Without augmenting its data, the VGG-16 model attained an accuracy of 91.7%. After training with augmented data, the model attained 97.9%. In evaluating the performance and accuracy of the classification of the deep transfer learning model, a confusion matrix was employed.