Scholarlink

<p>India boasts a wide array of timber species, encompassing a rich diversity characterised by distinct attributes, features, and uses. Identifying the type of timber is really a difficult task for a lay man. A subject-matter expert is always expected for this purpose. The proposed model aims to identify the origin of timber involving the grain structure. This can be achieved with the help of deep learning algorithms for creating a classification model on the type of timber species further a web application using Django network has been hosted through which the common name and type of timber can be analysed and presented to the user. A native dataset with timbers specific to southern region of Tamil Nadu state has been taken such as teak (Tectona grandis<b>),</b> neem (Azadirachta indica), pencil (Poinsettia) and vengai (Pterocarpus marsupium). A comparison of 3 different types of Convolutional Neural Network (CNN) model i.e., AlexNet, Lenet and LeNet with few shot learning is done. However LeNet few shot learning model seems to produce the highest accuracy among the other pretrained network, So LeNet model with few shot learning has been introduced for further processing. The cross-section of the timber is captured and uploaded in the web application through which the species and name of the timber is identified. The proposed work aims in creating a platform for ordinary people to understand the species of timber. The algorithm brought in to lime light has been evaluated using accuracy, F1 score, precision and recall values. Through various phases of training and testing with the profound minimal images the prototypical network was able to perform on the higher side with a score of 0.92 and 93% accuracy. The state of art models is nearer to performance but due to limited dataset the proposed few shot learning (prototypical network) architecture has shown a better performance. The web application on the other hand makes the interface easier to use in handy manner. Provided with the photograph of the cross section of the timber without the requirement of high-resolution image is another advantage of the proposed model which in turn increases the application perspective of the proposed work.</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

Wood species classification using prototypical networks: a few shot learning model

  • 期刊论文

  • 发表日期：

    2025年1月17日

查看全文
  • M. Diviya,
  • M. Subramanian

摘要

India boasts a wide array of timber species, encompassing a rich diversity characterised by distinct attributes, features, and uses. Identifying the type of timber is really a difficult task for a lay man. A subject-matter expert is always expected for this purpose. The proposed model aims to identify the origin of timber involving the grain structure. This can be achieved with the help of deep learning algorithms for creating a classification model on the type of timber species further a web application using Django network has been hosted through which the common name and type of timber can be analysed and presented to the user. A native dataset with timbers specific to southern region of Tamil Nadu state has been taken such as teak (Tectona grandis), neem (Azadirachta indica), pencil (Poinsettia) and vengai (Pterocarpus marsupium). A comparison of 3 different types of Convolutional Neural Network (CNN) model i.e., AlexNet, Lenet and LeNet with few shot learning is done. However LeNet few shot learning model seems to produce the highest accuracy among the other pretrained network, So LeNet model with few shot learning has been introduced for further processing. The cross-section of the timber is captured and uploaded in the web application through which the species and name of the timber is identified. The proposed work aims in creating a platform for ordinary people to understand the species of timber. The algorithm brought in to lime light has been evaluated using accuracy, F1 score, precision and recall values. Through various phases of training and testing with the profound minimal images the prototypical network was able to perform on the higher side with a score of 0.92 and 93% accuracy. The state of art models is nearer to performance but due to limited dataset the proposed few shot learning (prototypical network) architecture has shown a better performance. The web application on the other hand makes the interface easier to use in handy manner. Provided with the photograph of the cross section of the timber without the requirement of high-resolution image is another advantage of the proposed model which in turn increases the application perspective of the proposed work.