| 基于改进卷积神经网络的红外图像弱小目标检测方法研究 |
 点此下载全文 |
| 引用本文：唐亮，王海云，陈晓范.基于改进卷积神经网络的红外图像弱小目标检测方法研究[J].计算技术与自动化,2025,(3):100-105 |
| 摘要点击次数: 366 |
| 全文下载次数: 225 |
|
|
| 中文摘要:针对现有红外小目标检测精度低、虚警率高的现实问题，提出了一种基于改进U-Net的红外小目标检测方法。利用级联通道和空间注意模块捕获红外目标特征，实现渐进式特征交互和自适应特征增强，并将其应用特征金字塔融合模块提取不同尺度的特征信息。通过重复特征融合和增强，从而有效融合和充分利用小目标的上下文信息，提高模型处理和提取深层特征的能力。实验结果表明，与Transformer检测方法相比，所提方法IOU、POD分别提升1.16%、1.40%，FAR降低19.16%。实验结果验证了所提方法的有效性和实用性，该方法具有广阔的应用前景。 |
| 中文关键词:红外图像 小目标检测 卷积神经网络 特征提取 注意力 特征金字塔 |
| |
| Research on Infrared Image Weak Object Detection Method Based on Improved Convolutional Neural Network |
|
|
| Abstract:To address the issues of low accuracy and high false alarm rates in current infrared small target detection, a novel method based on an enhanced U-Net architecture is proposed. The method utilizes cascaded channels and spatial attention modules to capture infrared target features, enabling progressive feature interaction and adaptive feature enhancement. Additionally, it employs feature pyramid fusion modules to extract feature information across different scales. Through repeated feature fusion and enhancement, the contextual information of small targets is effectively integrated and fully leveraged, thereby enhancing the model’s capability to process and extract deep features. Experimental results demonstrate that, compared to the Transformer detection method, the proposed approach improves IOU and POD by 1.16% and 1.40% respectively, while reducing FAR by 19.16%. These findings confirm the effectiveness and practicality of the proposed method, which suggests its potential for broad application. |
| keywords:infrared images small object detection convolutional neural networks feature extraction attention feature pyramid |
| 查看全文 查看/发表评论 下载pdf阅读器 |
位访问者,今日访问
