基于改进SSD算法的遥感影像小目标快速检测

Small-object Fast Detection in Optical Remote Sensing Images Based on a Modified SSD

针对遥感影像小目标检测难度大、准确率低、耗时长等问题,本文提出一种基于改进SSD算法以提升遥感影像小目标实时检测精度的方法.(1)采用深度可分离卷积代替普通卷积层,从而减少计算量、加快目标检测速度;(2)修改SSD网络层数,最终使用7个卷积层的SSD作为检测器,选取其中4个卷积层的输出进行检测,进一步减少模型复杂度和训练难度;(3)修改了每个检测层所产生的候选框大小,提高检测精度.实验结果表明:所提出的模型平均准确率达到82.40%,平均每张影像检测耗时1.86 s,充分验证了该方法的有效性.本文提出的基于改进的SSD算法在遥感影像小目标检测中具备有效性和实时性,在遥感影像小目标检测任务中效果良好.

Aiming at the difficulty,accuracy and time consumption of small object detection in remote sensing image,a real-time detection method of small object in remote sensing image based on an improved SSD algorithm was proposed.(1) the depth-separable convolution is adopted to replace the ordinary convolution layer,so as to reduce the computation and speed up the target detection.(2) carefully modify the number of SSD network layers,finally use SSD of 7 convolutional layers as the detector,and select the output of 4 convolutional layers for detection,further reducing the complexity of the model and the difficulty of training.(3) carefully modified the anchor frame size and length-width ratio of each detection layer to improve the detection accuracy.The average accuracy of the proposed model reached 82.40%,and the average detection time per image was 1.86 s,which fully verified the effectiveness of the method.The proposed method in this paper is effective and real-time,and it has a good effect on the task of detecting small objects in remote sensing images.