针对目前遥感图像小目标检测任务中易出现漏检和误检的问题，提出一种SCS-YOLO［SMCA + CSC + SIoU（shape-aware intersection over union loss）-you only look once］的遥感图像小目标检测算法。首先，针对遥感图像中目标小而聚集的问题，构建空间多尺度卷积注意力（spatial multi-scale convolutional attention, SMCA），提升模型对空间和通道信息的特征提取能力；其次，针对深层网络传递时小目标语义信息容易丢失的问题，设计聚合亚像素卷积（concentrated sub-pixel convolution, CSC），采用多尺度聚合特征提取方法，增强了网络对语义信息的提取能力；最后，将SIoU损失函数替代原模型中的CIoU（complete intersection over union loss）损失函数，加快了网络的收敛速度。SCS-YOLO模型在RSOD和NWPU VHR-10数据集上，平均精确率的平均值（mAP）分别达到97%和90.9%，相较于原模型分别提升了2.2%和2.7%，可见该方法在遥感图像小目标检测任务中的有效性。

In order to solve the problems of missed detection and false detection in the current remote sensing image small target detection task, a SMCA + CSC + shape-aware intersection over union loss (SIoU)-you only look once (SCS-YOLO) remote sensing image small target detection algorithm was proposed. Firstly, in response to the problem of small and clustered targets in remote sensing images, a spatial multi-scale convolutional attention module(SMCA) was constructed to improve the model's feature extraction ability of spatial and channel information. Secondly, in order to solve the problem that the semantic information of small targets was easy to be lost during deep network transmission, the aggregation subpixel convolution module concentrated sub-pixel convolution (CSC) was designed, and the multi-scale aggregation feature extraction method was used to enhance the ability of the network to extract semantic information. Finally, the SIoU loss function was used to replace the complete intersection over union loss(CIoU) loss function in the original model, which accelerated the convergence speed of the network. The average of the average precision(mAP) of the SCS-YOLO model reaches 97% and 90.9% on the RSOD and NWPU VHR-10 datasets, respectively, which is 2.2% and 2.7% higher than that of the original model, which shows the effectiveness of the method in the small target detection task of remote sensing images.