| 基于改进YOLOv8的圆形合作目标检测定位算法研究 |
| 投稿时间：2024-02-08 修订日期：2024-03-08 点此下载全文 |
| 引用本文： |
| 摘要点击次数: 77 |
| 全文下载次数: 0 |
|
| 基金项目:吉林省科技发展计划重点研发项目（20200401063GX） |
|
| 中文摘要:针对视觉测量中圆形合作目标面对低照度或复杂背景下导致合作目标识别精度较低或定位能力较差等问题，本文提出了一种基于卷积神经网络（Convlutional Neural Networks,CNNs）对YOLOv8算法进行优化设计的模型。本文设计的模型共有225层网络，约300万参数，8.2G FLOPs计算量。采用不同条件下的圆形合作目标数据集进行模型训练，同时在训练过程中实时监控模型的性能指标和计算效率，对模型进行细致调整和优化。实验结果表明，本文算法具有99%准确率，92%召回率，92%平均精度，相较于霍夫变换、YOLOv3等传统特征提取方法，分别在精确率提升了14%、4%；召回率提升17%、2%；平均精度提升10%、2%。本文算法可以在低照度环境、复杂背景或目标形状微小变化等多变条件下，显著提高圆形合作目标的识别定位精度。 |
| 中文关键词:视觉测量 YOLOv8 精确率 召回率 平均精度 |
| |
| Research on Circular Cooperative Object Detection and Localization Algorithm Based on Improved YOLOv8 |
|
|
| Abstract:This paper proposes a model based on Convolutional Neural Networks (CNNs) for optimizing the YOLOv8 algorithm to address the issues of low recognition accuracy or poor localization ability of circular cooperative targets in visual measurement under low illumination or complex backgrounds. The model designed in this article has a total of 225 layers of network, approximately 3 million parameters, and 8.2G FLOPs computation. Train the model using circular cooperative target datasets under different conditions, while monitoring the performance indicators and computational efficiency of the model in real-time during the training process, and making detailed adjustments and optimizations to the model. The experimental results show that the algorithm proposed in this paper has 99% accuracy, 92% recall, and 92% average accuracy. Compared with traditional feature extraction methods such as Hough transform and YOLOv3, it has improved accuracy by 14% and 4%, respectively; 17% and 2% increase in recall rate; Average accuracy increased by 10% and 2%. The algorithm proposed in this article can significantly improve the recognition and localization accuracy of circular cooperative targets under various conditions such as low illumination, complex backgrounds, or small changes in target shape. |
| keywords:vision measurement YOLOv8 precision recall mean average precision |
| HTML 查看/发表评论 下载PDF阅读器 |