基于不同卷积神经网络的目标检测算法对比研究

Comparative study of target detection algorithms based on different convolution neural networks

为实现基于遥感影像的地物信息智能化、快速化、高精度的目标检测和提取,本文基于卷积神经网络理论,对R-CNN、SPPNet、 Fast R-CNN、Faster R-CNN及Mask R-CNN 5种不同卷积神经网络算法进行对比研究。结果表明,Mask R-CNN算法在分类精度和标定框精度上具有明显优势,运行时间较传统卷积神经网络算法有较大提升,通过检测获得的二值Mask为提取影像建筑物轮廓等后续工作提供技术准备,可在高分辨率遥感影像地物检测提取中应用。

In order to achieve intelligent,rapid,and high precision target detection and extraction of remote sensing image ground information,this paper compares five different convolutional neural network algorithms based on convolutional neural network theory,including R-CNN,SPPNet,Fast R-CNN,Faster R-CNN,and Mask R-CNN.The results show that the Mask R CNN algorithm has significant advantages in classification accuracy and calibration box accuracy,and its running time is significantly improved compared to traditional convolutional neural network algorithms.The binary Mask obtained through detection can also provide technical preparation for subsequent work such as extracting building contours from images,and can be reasonably applied in high resolution remote sensing image feature detection and extraction.