一种基于深度脉冲神经网络的高帧率目标检测方法
By transforming, normalizing, and compressing tensors in a deep spiking neural network, a lightweight, high-frame-rate target detection model is constructed, which solves the problems of high computational complexity and low accuracy in small target detection in existing technologies, and achieves low-power, high-efficiency target detection.
CN117830617BActive Publication Date: 2026-07-17SHANGHAI AEROSPACE CONTROL TECH INST
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- SHANGHAI AEROSPACE CONTROL TECH INST
- Filing Date
- 2023-12-26
- Publication Date
- 2026-07-17
AI Technical Summary
Technical Problem
Existing high frame rate target detection technologies have high computational complexity, large computational overhead, and low accuracy in detecting small-scale targets, making it difficult to achieve low-power and efficient detection on mobile devices.
Method used
A lightweight, high-frame-rate target detection model is constructed by employing a deep spiking neural network, using YOLOv3-Tiny transformation, channel normalization, threshold-imbalanced signed neurons, and STDP unsupervised learning, combined with tensor decomposition and quantization compression.
Benefits of technology
It achieves low power consumption, high precision, and high real-time target detection, is suitable for mobile devices, and reduces computational complexity and energy consumption.
✦ Generated by Eureka AI based on patent content.
Smart Images
Figure CN117830617B_ABST
Abstract
本发明涉及一种基于深度脉冲神经网络的高帧率目标检测方法，由YOLOv3‑Tiny神经网络模型转换得到脉冲神经网络；在脉冲神经网络每层的通道维度上，通过最大激活值对权值及偏置进行归一化处理；使用阈值不平衡的有符号神经元方法设置神经元的正负激活阈值，选择脉冲神经网络中需要被激活的神经元；基于STDP的无监督学习训练方法，对优化后的脉冲神经网络进行训练，得到高帧率目标检测模型；对高帧率目标检测模型进行张量分解与量化压缩，实现轻量化处理；将待检测图像输入轻量化的高帧率目标检测模型，输出图像中目标类别和目标边界位置坐标。本发明实现了高精度、低功耗、高实时性的目标检测，减小了模型参数量、模型大小及模型训练时间。
Need to check novelty before this filing date? Find Prior Art