Target detection method and system based on FPGA lightweight CNN network

By deploying a lightweight CNN network based on FPGA on the intelligent inspection robot, the problem of cloud recognition latency was solved, enabling real-time and accurate detection of building materials while reducing computation and resource consumption.

CN118840727BActive Publication Date: 2026-06-19CHINA UNIV OF GEOSCIENCES (WUHAN)
2 Cites 0 Cited by

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
CHINA UNIV OF GEOSCIENCES (WUHAN)
Filing Date
2024-07-24
Publication Date
2026-06-19

AI Technical Summary

⚠Technical Problem

When the intelligent inspection robot uses deep learning inference algorithms on the cloud host to identify the quantity of pallets and building materials on them, the image transmission rate is unstable, resulting in excessively long recognition delays.

⚗Method used

Design a lightweight CNN network based on FPGA, combining YOLOv4 and MobileNetV2 models, and deploy it on an embedded FPGA development board to achieve real-time target detection through Batch Normalization (BN) fusion network compression, model pruning, and quantization.

🎯Benefits of technology

Real-time and accurate building material detection is achieved on embedded devices, reducing computation and resource consumption, and the detection rate is comparable to that of a PC GPU, with the advantages of low power consumption and small size.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure 1
    Figure 1
  • Figure 2
    Figure 2
Patent Text Reader

Abstract

This application provides a target detection method and system based on a lightweight CNN network using FPGA, relating to the field of target detection. The method includes: determining the performance indicators of an inspection robot; designing a lightweight convolutional neural network by combining a YOLOv4 model and a MobileNetV2 network; compressing the lightweight convolutional neural network using a Batch Normalization (BN) fusion network; acquiring and preprocessing the dataset; pre-training the neural network model; optimizing the pre-trained lightweight convolutional neural network using a model pruning strategy; quantizing the optimized lightweight convolutional neural network and deploying it to an embedded platform; and using images acquired by the inspection robot, combined with the deployed lightweight convolutional neural network, to perform building material detection, thus completing the rapid inspection task of the intelligent inspection robot. The lightweight convolutional neural network is designed to address the problem of limited memory resources on embedded platforms.
Need to check novelty before this filing date? Find Prior Art