结合高效通道注意力的轻量级遥感影像目标检测方法

Lightweight Remote Sensing Image Target Detection Method Combined with Efficient Channel Attention

针对常规遥感影像目标检测模型难以在机载、星载等低算力场景下部署的问题,本文提出一种轻量级遥感影像目标检测模型。模型通过Ghost特征提取模块与高效通道注意力机制组成轻量级骨干网络来进行特征提取与筛选,然后将获取到的特征图送入融合金字塔中生成3张语义特征更为丰富的特征图参与多尺度目标检测。在多源混合数据集上的测试结果表明,本文模型对各类别目标的检测精度均优于其余对比模型,对不同场景下的目标具有良好的泛化能力。训练后模型占用内存小,推理参数量低,在低算力的测试场景下也能够进行实时检测,适合部署于算力较低的边缘计算场景。

Aiming at the problem that the conventional remote sensing image target detection model is difficult to deploy in airborne,spaceborne and other low computational scenes,a lightweight remote sensing image target detection model is proposed.The model uses ghost feature extraction module and efficient attention mechanism to form a lightweight backbone network to realize feature extraction and screening,and then sends the obtained feature map into the fusion pyramid to generate three feature maps with richer semantic features to participate in multi-scale target detection.The test results on multi-source mixed data sets show that the model in this paper is superior to other comparison models in the detection accuracy of each category,and has good generalization ability for targets in different scenarios.After training,the model occupies less memory and has low reasoning parameters.It can also carry out real-time detection in low computing power test scenarios,which is suitable for deployment in low computing power edge computing scenarios.