Scholarlink

<p>We present a method for analyzing the decision boundary in deep learning models, focusing on the study of the minimum distance of images to this boundary and the evolution of this margin throughout the training of a convolutional neural network. To enhance the robustness of the model, we generate, in addition to the original images, a perturbed version of each image while preserving the same label. The dataset used includes 15 types of otoliths, calcified structures located in the inner ear of fish. Recognizing their shape provides economically and ecologically relevant information about the status of fish populations worldwide. Indeed, otolith analysis allows researchers to determine species stocks, estimate fish age, and study taxonomy. However, identifying species from otolith images remains a labor-intensive task requiring specialized expertise and considerable resources. Recent approaches to otolith image classification mainly rely on deep neural networks, whose primary challenge is generalizability – the ability to maintain high performance on unseen data. Our study aims to better understand this issue by evaluating the decision boundary throughout the training process. Our results show that the convolutional neural network we developed achieves an accuracy of 92% for classifying the 15 otolith types. We also demonstrate that the distance between images and the decision boundary is significantly larger under adversarial training than under standard training, highlighting the effectiveness and robustness of our approach. To further validate the robustness and generalization capability of this method, we conducted similar experiments on the MNIST and Fashion MNIST datasets, obtaining comparable results that confirm the model trained with adversarial examples is more robust and confident in its predictions.</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

Efficient Model Based Decision Boundary to Improve Otolith Classification Using Deep Neural Networks

  • 期刊论文

  • 发表日期：

    2026年5月28日

查看全文
  • Mohamed Ouriha,
  • Yousef El Mourabit,
  • Youssef El Habouz,
  • Omar El Mansouri,
  • Fathallah Nouboud,
  • Boujemaa Nassiri,
  • Otman Maarouf

摘要

We present a method for analyzing the decision boundary in deep learning models, focusing on the study of the minimum distance of images to this boundary and the evolution of this margin throughout the training of a convolutional neural network. To enhance the robustness of the model, we generate, in addition to the original images, a perturbed version of each image while preserving the same label. The dataset used includes 15 types of otoliths, calcified structures located in the inner ear of fish. Recognizing their shape provides economically and ecologically relevant information about the status of fish populations worldwide. Indeed, otolith analysis allows researchers to determine species stocks, estimate fish age, and study taxonomy. However, identifying species from otolith images remains a labor-intensive task requiring specialized expertise and considerable resources. Recent approaches to otolith image classification mainly rely on deep neural networks, whose primary challenge is generalizability – the ability to maintain high performance on unseen data. Our study aims to better understand this issue by evaluating the decision boundary throughout the training process. Our results show that the convolutional neural network we developed achieves an accuracy of 92% for classifying the 15 otolith types. We also demonstrate that the distance between images and the decision boundary is significantly larger under adversarial training than under standard training, highlighting the effectiveness and robustness of our approach. To further validate the robustness and generalization capability of this method, we conducted similar experiments on the MNIST and Fashion MNIST datasets, obtaining comparable results that confirm the model trained with adversarial examples is more robust and confident in its predictions.