Scholarlink

The paper proposes a scheme to design a hardware convolutional layer as part of hardware object detection systems like convolutional neural networks (CNN), region convolutional neural networks (RCNN), single-shot multi-box detectors (SSD), and You Only Look Once (YOLO). Hardware-based object detection systems play a crucial role in machine vision techniques, as they can be easily integrated with robot circuitry. A system with a soft-core processor is developed and downloaded on the SP 6E slice to create the hardware system media, which is adapted to function as a hard convolutional layer. External memories, such as EEPROM and synchronous dynamic RAM with dual data rate, were attached to the system via the processor local bus to meet intensive computation requirements. An Ethernet IP core is attached to the system to act as a data acquisition remote port. The designed configuration’s performance is tested and compared with MATLAB software. The error was lower than 0.002%.

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

Convolutional Layer Unit for FPGA-Based Object Detection Systems

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Mazin Rejab Khalil,
  • Ezideen A. Hasso,
  • Qusay Hamid Ali

摘要

The paper proposes a scheme to design a hardware convolutional layer as part of hardware object detection systems like convolutional neural networks (CNN), region convolutional neural networks (RCNN), single-shot multi-box detectors (SSD), and You Only Look Once (YOLO). Hardware-based object detection systems play a crucial role in machine vision techniques, as they can be easily integrated with robot circuitry. A system with a soft-core processor is developed and downloaded on the SP 6E slice to create the hardware system media, which is adapted to function as a hard convolutional layer. External memories, such as EEPROM and synchronous dynamic RAM with dual data rate, were attached to the system via the processor local bus to meet intensive computation requirements. An Ethernet IP core is attached to the system to act as a data acquisition remote port. The designed configuration’s performance is tested and compared with MATLAB software. The error was lower than 0.002%.