Scholarlink

<p>The persistence of Lumpy Skin Disease (LSD) presents numerous challenges in its diagnosis and management; hence, better and more robust predictive models are needed. Existing deep learning techniques may require voluminous data with labeled datasets and remain susceptible to adversarial attacks. In addition, these models do not exhibit interpretability, a critical feature in veterinary applications where interpretability can greatly affect trust and usability. In our proposed work, we introduced a new method using fine-grained feature fusion (FGFF), adaptive transfer learning, uncertainty-aware active learning (UAAL), longitudinal temporal fusion (LTF), and robust interpretable model (RIM) techniques to improve disease prediction and management. Compared with traditional deep learning methods, FGFF combines handcrafted and deep learning features that increase the classification accuracy by an additional 5%, as it captures disease characteristics that are not evident in pretrained networks. Furthermore, adaptive transfer learning (ATL) expands the same method by adapting pretrained models to disease-specific datasets through fine-tuning and selective layer freezing to achieve an additional accuracy of 10%. Moreover, UAAL reduces the required annotations by approximately 30% and improves the model efficacy through uncertainty metrics employed to select informative samples. The LTF also captures temporal disease progression via advanced recurrent networks with an accuracy of 80% and forecasts the future disease state. Finally, RIM helps to build a robust model by employing adversarial training and ensuring the interpretability of the model in terms of attention mechanisms concerning robustness against adversarial attacks, which amounts to 95%. All these combined methodologies not only outperform older predictive models in terms of accuracy, robustness, interpretability and efficient data usage but also set a new benchmark in the development of diagnostic and prognostic tools in animal healthcare. This fusion approach breaks new ground for the development of diagnostic and prognostic tools in animal healthcare, especially for managing epidemics such as LSD. Futher the novelty of the model includes FGFF which is used here to optimize the level of predicting LSD using detailed data integration where as ATL futher used to cope up with particular special features of LSD. The Data fusion granularity helps to indentify early stages of LSD which makes it easier to have some corrective measures accordingly.</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

Design of an Iterative Model for Lumpy Skin Disease Prediction Using Fine-grained Feature Fusion and Adaptive Transfer Learning

  • 期刊论文

  • 发表日期：

    2025年12月17日

查看全文
  • Sandeep Sharma,
  • Kapil Joshi,
  • Rajesh Singh

摘要

The persistence of Lumpy Skin Disease (LSD) presents numerous challenges in its diagnosis and management; hence, better and more robust predictive models are needed. Existing deep learning techniques may require voluminous data with labeled datasets and remain susceptible to adversarial attacks. In addition, these models do not exhibit interpretability, a critical feature in veterinary applications where interpretability can greatly affect trust and usability. In our proposed work, we introduced a new method using fine-grained feature fusion (FGFF), adaptive transfer learning, uncertainty-aware active learning (UAAL), longitudinal temporal fusion (LTF), and robust interpretable model (RIM) techniques to improve disease prediction and management. Compared with traditional deep learning methods, FGFF combines handcrafted and deep learning features that increase the classification accuracy by an additional 5%, as it captures disease characteristics that are not evident in pretrained networks. Furthermore, adaptive transfer learning (ATL) expands the same method by adapting pretrained models to disease-specific datasets through fine-tuning and selective layer freezing to achieve an additional accuracy of 10%. Moreover, UAAL reduces the required annotations by approximately 30% and improves the model efficacy through uncertainty metrics employed to select informative samples. The LTF also captures temporal disease progression via advanced recurrent networks with an accuracy of 80% and forecasts the future disease state. Finally, RIM helps to build a robust model by employing adversarial training and ensuring the interpretability of the model in terms of attention mechanisms concerning robustness against adversarial attacks, which amounts to 95%. All these combined methodologies not only outperform older predictive models in terms of accuracy, robustness, interpretability and efficient data usage but also set a new benchmark in the development of diagnostic and prognostic tools in animal healthcare. This fusion approach breaks new ground for the development of diagnostic and prognostic tools in animal healthcare, especially for managing epidemics such as LSD. Futher the novelty of the model includes FGFF which is used here to optimize the level of predicting LSD using detailed data integration where as ATL futher used to cope up with particular special features of LSD. The Data fusion granularity helps to indentify early stages of LSD which makes it easier to have some corrective measures accordingly.