基于奇异值分解的航空遥感图像小目标提取方法设计

Design of Small Target Extraction Method for Aerial Remote Sensing Images Based on Singular Value Decomposition

摘要:

航空遥感图像具有目标尺寸小且信息量丰富的特点。然而,由于受到低秩噪声的干扰,这导致了小目标的提取精确度下降和漏检现象增多。为了解决这一问题,提出了一种基于奇异值分解的航空遥感图像小目标提取方法。依据奇异值分解准则,结合不均匀变化奇异值特征向量提取小目标,计算奇异值变换能量增益。在此约束条件下构建信号空间杂波的协方差矩阵,反映信号的分布情况以及不同信号之间的联系。使用奇异值分解矩阵,避免因计算杂波协方差矩阵带来的影响,反映图像小目标的形状、大小、纹理等信息。奇异值分解图像矩阵,反映图像中行、列、像素强度信息。对正交矩阵分解,根据选择新的基向量组,重建图像矩阵。压缩图像,将其分为分散目标、完全叠加目标、部分叠加目标三个部分,计算这三个部分的能量衰减倍数,完成航空遥感图像小目标提取。由实验结果可知,该技术的召回率和准确率高,且“船”小目标漏检最大量为4只,说明使用该技术具有精准、高效提取效果。

Abstract:

Aerial remote sensing images have the characteristics of small target size and rich information content. However, due to the interference of low rank noise, this leads to a decrease in the accuracy of extracting small targets and an increase in missed detections. To address this issue, a small target extraction method for aerial remote sensing images based on singular value decomposition is proposed. According to the singular value decomposition criterion, combined with the non-uniform variation of singular value eigenvectors, small targets are extracted and the energy gain of singular value transformation is calculated. Construct a covariance matrix of signal spatial clutter under this constraint, reflecting the distribution of signals and the connections between different signals. Using singular value decomposition matrix to avoid the impact of calculating clutter covariance matrix, reflecting the shape, size, texture and other information of small targets in the image. Singular value decomposition image matrix, reflecting the strength information of rows, columns, and pixels in the image. Decompose the orthogonal matrix and reconstruct the image matrix by selecting a new set of basis vectors. Compress the image and divide it into three parts: scattered targets, fully stacked targets, and partially stacked targets. Calculate the energy attenuation factor of these three parts to complete the extraction of small targets in aerial remote sensing images. According to the experimental results, the recall and accuracy of this technology are high, and the maximum number of missed small targets on the ship is 4, indicating that the use of this technology has a precise and efficient extraction effect.