Scholarlink

In agriculture, environmental monitoring, and biodiversity conservation, plant identification is essential. But the task is hard, especially when there aren't many labelled data to train deep learning models with. In this study, we provide a novel method for plant identification tasks with less data that combines MobileNet-V2 architecture and transfer learning. Compared to training from scratch, we show notable gains in classification accuracy by optimising pre-trained MobileNet-V2 on a limited dataset. We carry out in-depth tests on a variety of plant datasets to show the efficacy and efficiency of our methodology. Additionally, we examine how well MobileNet-V2's learnt features translate to other plant species and datasets. Our results show that MobileNet-V2 and transfer learning together have great potential as a solution for plant identification problems with a small amount of labelled data. This could lead to useful applications in the fields of ecology, agriculture, and environmental sciences The results highlight the potential of MobileNet-V2 transfer learning as a useful approach for goal-oriented plant identification tasks with little labelled data, with applications in environmental sciences, ecology, and agriculture.

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

Leveraging Transfer Learning for Plant Identification with Limited Data: A MobileNet-V2 Based Approach

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Himanshu Gupta,
  • Himanshu Kumar Diwedi,
  • Esha Tripathi,
  • Abhay Kumar Tripathi

摘要

In agriculture, environmental monitoring, and biodiversity conservation, plant identification is essential. But the task is hard, especially when there aren't many labelled data to train deep learning models with. In this study, we provide a novel method for plant identification tasks with less data that combines MobileNet-V2 architecture and transfer learning. Compared to training from scratch, we show notable gains in classification accuracy by optimising pre-trained MobileNet-V2 on a limited dataset. We carry out in-depth tests on a variety of plant datasets to show the efficacy and efficiency of our methodology. Additionally, we examine how well MobileNet-V2's learnt features translate to other plant species and datasets. Our results show that MobileNet-V2 and transfer learning together have great potential as a solution for plant identification problems with a small amount of labelled data. This could lead to useful applications in the fields of ecology, agriculture, and environmental sciences The results highlight the potential of MobileNet-V2 transfer learning as a useful approach for goal-oriented plant identification tasks with little labelled data, with applications in environmental sciences, ecology, and agriculture.