计算机科学
红外线的
人工智能
联营
卷积(计算机科学)
特征提取
特征(语言学)
连接(主束)
边缘检测
计算机视觉
编码器
噪音(视频)
图像(数学)
模式识别(心理学)
跳跃
卷积神经网络
目标检测
GSM演进的增强数据速率
解码方法
特征检测(计算机视觉)
图像处理
探测理论
可分离空间
降噪
信噪比(成像)
语义学(计算机科学)
算法
跟踪(教育)
作者
Jianjun Ni,Guangyi Tang,Yonghao Zhao,Yang Gu,Weidong Cao
标识
DOI:10.1109/lgrs.2025.3609487
摘要
The detection of small targets from infrared images with a low signal-to-noise ratio and complex backgrounds has important practical significance in applications such as maritime rescue and traffic management. However, due to factors such as low image resolution, small-target size, and lack of prominent features, infrared targets are easily submerged in the background containing noise and clutter. To deal with these problems, a U-Net-based method with a double-bridge jump connection for infrared small-target detection is proposed in this article. First, the encoder expands the single-channel infrared image to 16 channels and then uses a depthwise separable asymmetric convolutional network to improve the extraction ability of small infrared targets. Then, a double-bridge jump connection is designed to integrate low-level detail information into high-level semantic information. One of them uses a fully connected layer, and another introduces an edge detection module (EDM) that can improve detection accuracy. In addition, the decoder can effectively solve the problem of feature information loss caused by pooling or convolution through dense upsampling. Finally, experiments are conducted, and the results show that the proposed method can effectively detect infrared small targets and outperform state-of-the-art methods in various detection indicators.
科研通智能强力驱动
Strongly Powered by AbleSci AI