Scholarlink

<p>The classification of plant diseases poses a significant challenge in smart agriculture. Most of the few-shot learning methods for classifying plant diseases currently rely on metrics like Euclidean distance or cosine similarity to model marginal feature distributions. However, these metrics often fail to capture the intricate and non-linear relationships inherent in fine-grained disease manifestations. To address this limitation, we propose an innovative framework named Spatial Feature Fusion with Brownian Distance Covariance (SFF-BDC). To our knowledge, this is the first work to incorporate joint distribution modeling into this field. The framework consists of two complementary modules: a deep Brownian Distance Covariance (BDC) metric that effectively measures various statistical dependencies between query and support samples, and a novel Spatial Feature Fusion (SFF) module that enriches feature representations by explicitly incorporating directional spatial information, which optimizes the input for subsequent BDC computation. Evaluations on various splits of the public PlantVillage dataset show remarkable accuracies ranging from 79.28% to 93.43%. Specifically, the framework’s superior performance in challenging, fine-grained classification scenarios strongly validates our core hypothesis.</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

A few-shot learning approach based on Brownian distance covariance for plant disease classification

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2025年12月11日

查看全文
  • Chunmao Li,
  • Fang Liu,
  • Li Xu,
  • Lina Zhang,
  • Xiaowei Jiang

摘要

The classification of plant diseases poses a significant challenge in smart agriculture. Most of the few-shot learning methods for classifying plant diseases currently rely on metrics like Euclidean distance or cosine similarity to model marginal feature distributions. However, these metrics often fail to capture the intricate and non-linear relationships inherent in fine-grained disease manifestations. To address this limitation, we propose an innovative framework named Spatial Feature Fusion with Brownian Distance Covariance (SFF-BDC). To our knowledge, this is the first work to incorporate joint distribution modeling into this field. The framework consists of two complementary modules: a deep Brownian Distance Covariance (BDC) metric that effectively measures various statistical dependencies between query and support samples, and a novel Spatial Feature Fusion (SFF) module that enriches feature representations by explicitly incorporating directional spatial information, which optimizes the input for subsequent BDC computation. Evaluations on various splits of the public PlantVillage dataset show remarkable accuracies ranging from 79.28% to 93.43%. Specifically, the framework’s superior performance in challenging, fine-grained classification scenarios strongly validates our core hypothesis.