欢迎访问西安知识产权运营服务平台
摘要
【 中文摘要 】
本发明公开了一种空中红外弱小目标检测方法、系统及计算机存储介质，方法包括：确定可学习候选框；生成目标图像中与可学习候选框对应的可学习候选特征；将目标图像输入骨干网络中，骨干网络采用加权双向循环特征金字塔网络，其包含可切换空洞卷积模块，骨干网络用于提取目标图像中不同分辨率的特征，并对提取的特征进行融合，获得融合特征；将可学习候选特征和融合特征输入动态检测模块，提取每个可学习候选框中的区域特征；对区域特征进行分类预测，获得检测结果。本发明采用加权双向循环特征金字塔网络，提高特征提取能力，并引入可切换空洞卷积，在较少图像信息丢失的情况下增大感受野，提高检测准确率。
【 英文摘要 】
The invention discloses a method, system and computer storage medium for detecting weak and small infrared targets in the air. The method includes : determining a learnable candidate frame; generating learnable candidate features corresponding to the learnable candidate frame in the target image; input the target image into the backbone network. The backbone network uses a weighted bidirectional cyclic feature pyramid network, which includes a switchable dilated convolution module. The backbone network is used to extract features of different resolutions in the target image and fuse the extracted features to obtain Fusion features; input the learnable candidate features and fusion features into the dynamic detection module, and extract the regional features in each learnable candidate box; classify and predict regional features to obtain detection results. According to the invention, a weighted bidirectional cyclic feature pyramid network is adopted, so that the feature extraction capability is improved; and switchable dilated convolution is introduced, so that the receptive field is increased under the condition of less image information loss, and the detection accuracy is improved
其他著录项
| 代理机构 | 西安鼎迈知识产权代理事务所(普通合伙) 61263 |
| 代理人 | 刘喜保 |
| 申请语言 | 汉语 |
| 审查员 | 白雪慧 |
其中，变换函数S(x)由一个5×5的平均池化层和一个1×1的卷积层构成，空洞率r默认设置为3，
和
采用锁定机制共享权重w，对于空洞率为r的卷积层增加额外的可训练权重
；
其中，变换函数S(x)由一个5×5的平均池化层和一个1×1的卷积层构成，空洞率r默认设置为3，
和
采用锁定机制共享权重w，对于空洞率为r的卷积层增加额外的可训练权重
；