Scholarlink

Quantization of deep neural networks has emerged as an enabler for deploying these models on resource-constrained edge devices. This paper evaluates current quantization techniques using a custom dataset for agricultural applications, specifically for detecting sugarbeet plants and weeds from an aerial perspective. The experimental results demonstrate significant improvements in inference speed with minimal accuracy loss whilst achieving state-of-the-art compression rates for both ResNet18 and ResNet50 architectures.

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

Comparing Quantization Techniques for DNNs in Precision Agriculture

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Domenic Drechsel,
  • Ali Ehteshami Bejnordi,
  • Stefan Henkler,
  • Kristian Rother

摘要

Quantization of deep neural networks has emerged as an enabler for deploying these models on resource-constrained edge devices. This paper evaluates current quantization techniques using a custom dataset for agricultural applications, specifically for detecting sugarbeet plants and weeds from an aerial perspective. The experimental results demonstrate significant improvements in inference speed with minimal accuracy loss whilst achieving state-of-the-art compression rates for both ResNet18 and ResNet50 architectures.