Scholarlink

At present, thousands of tree species have been described in the planet where there are countries with great diversity, for this, their identification and classification is a concise activity for industrial, economic or environmental purposes. However, this task is done visually or chemically, depending on the skill and techniques of the specialist. We consider it urgent to have a tool to automate this task by means of recent pattern recognition and deep learning techniques. In this paper we present a comparison of two proposals for classification by means of artificial vision, using convolutional neural networks (CNN), and Transfer Learning with Fine-Tunning. We started from the use of the VGG16 network and our own database of images of 39 Mexican timber species that we are making available to the public. In the first experimental stage, the model achieved 100% recognition using the original database. In the second test, performed to verify the stability of the results, the accuracy obtained was 99.48% with the original data.

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

Characterization and Classification of Mexican Woods Through Local Texture Analysis Using Deep Learning Techniques

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Juan Pablo Garduza-Ventura,
  • Juan Villegas-Cortez,
  • Arturo Zúñiga-López,
  • Román Anselmo Mora-Gutiérrez,
  • Josefa Díaz-Álvarez,
  • Francisco Chávez de la O

摘要

At present, thousands of tree species have been described in the planet where there are countries with great diversity, for this, their identification and classification is a concise activity for industrial, economic or environmental purposes. However, this task is done visually or chemically, depending on the skill and techniques of the specialist. We consider it urgent to have a tool to automate this task by means of recent pattern recognition and deep learning techniques. In this paper we present a comparison of two proposals for classification by means of artificial vision, using convolutional neural networks (CNN), and Transfer Learning with Fine-Tunning. We started from the use of the VGG16 network and our own database of images of 39 Mexican timber species that we are making available to the public. In the first experimental stage, the model achieved 100% recognition using the original database. In the second test, performed to verify the stability of the results, the accuracy obtained was 99.48% with the original data.