基于HR-BIT神经网络的图像变化检测

Research on image change detection based on HR-BIT neural network

针对现有的遥感图像变化检测方法准确率不高的问题,通过对HRNet网络结构进行优化,提出HR-BIT网络模型。该模型在特征提取的过程中融合多尺度特征信息,提高检测精度。引入连通和邻域的概念,通过对分类结果阈值处理消除孔洞,提高模型的准确率。在LEVIR-CD数据集上的实验结果表明,该模型有效地提高了图像变化检测的准确性。

Aiming at the low accuracy of existing remote sensing image change detection methods,this paper proposes a HR-BIT network model by optimizing the HRNet network structure.In the process of feature extraction,the multi-scale feature information is fused to improve the detection accuracy.The concepts of connectivity and neighborhood are introduced,and the holes are eliminated by processing the threshold of classification results to improve the accuracy of the model.The experimental results on LEVIR-CD data set show that this model effectively improves the accuracy of image change detection.