Abstract:

The problem of target detection in remote sensing images has always been a hot and difficult point in the field of remote sensing image processing. Traditional detection algorithms have low performance when solving complex scenes with large scale differences. However, it is difficult to balance the accuracy and real-time performance of remote sensing targets using deep learning. In response to this problem, this paper designs a lightweight network that uses multi-scale fusion feature detection targets, and proposes an attention mechanism that can generate pixel adaptive feature weights from three dimensions to help extract salient features. The latest optimization algorithm is used to improve the performance of the model, while reducing the amount of calculations and ensuring the detection accuracy. Experimental results show that the model MAP@0.5 can reach 0.945 and F1 can reach 0.841, and the detection speed can meet the real-time requirements.

Key words: pattern recognition, remote sensing image, target detection, attention mechanism, multi-scale, optimization algorithm

摘要：

遥感影像中目标的检测问题一直是遥感图像处理领域的热点和难点。传统的检测算法，在解决场景复杂，尺度差异大的目标时性能不高，而使用深度学习很难兼顾遥感目标的准确性和实时性。针对这一问题，设计了一种利用多尺度融合特征检测目标的轻量级网络，并提出一种能够从三个维度上生成像素自适应特征权重的注意力机制帮助提取显著特征，同时采用了最新的优化算法改善模型的性能，在减少计算量的同时保证了检测精度。实验结果表明，该模型MAP@0.5可达0.945，F1可达0.841，检测速度满足实时性要求。

关键词: 模式识别, 遥感影像, 目标检测, 注意力机制, 多尺度, 优化算法