高分辨率特征增强图像的小目标检测算法

A Small Target Detection Algorithm for High⁃resolution Feature Enhanced Images

在高分辨率特征增强图像小目标特征检测过程中,局部特征与全局目标识别参量之间的参量分布不均,无法在目标图像区域内识别有效特征信息,导致实际检测结果无法满足应用要求。为了解决这一问题,利用鲁棒鉴别特征学习完成高分辨率特征增强图像的小目标检测。该算法中的D3Tes2Block结构单元提取小目标的多尺度特征,并通过损失函数优化,降低特征损失以提高对小目标区域的关注度,在鲁棒性特征学习结果的基础上,根据学习尺度参量进行目标区检测信息聚合,输出小目标检测结果。实验结果显示,本文方法能够有效提升小目标检测精准度,且整体性能输出稳定,更加适合实际场景的应用。

In the process of small target feature detection in high‐resolution feature‐enhanced images,the uneven distri‐bution of parameters between local features and global target recognition parameters makes it difficult to identify effective feature information within the target image area,resulting in actual detection results unable to meet application require‐ments.To tackle this issue,robust discriminative feature learning was utilized to achieve small target detection in high‐reso‐lution feature‐enhanced images.The D3 Tes2Block structural unit in this algorithm extracts multi‐scale features of the small targets,and through the loss function optimization,reduces feature loss to improve the focus on the small target area.Based on the robust feature learning results,target area detection information is aggregated according to the learning scale parameters to output small target detection results.The experimental results showed that the proposed method can effec‐tively improve the accuracy of small object detection,and the overall performance output is stable,which is more suitable for practical application scenes.