Min Wang

Existing deep learning supervised change detection (CD) networks still have room for improvement, as they do not fully utilize multiscale features in their feature extraction, resulting in insufficient feature representation ability and edge blurring problems of the constructed CD networks. In this paper, we proposed a very high-resolution (VHR) remote sensing image CD network (ACMFNet) with an asymmetric convolution residual block (ACRB) and multiscale fusion (MSF) to improve the feature representation ability of the CD network and alleviate the edge blurring problem. ACMFNet consists of two main subnetworks: an ACRB feature extraction encoder and an MSF decoder. The asymmetric convolution residual block (ACRB) is constructed based on asymmetric convolution and focuses on extraction at the edges of the features and is more robust to rotations, flip distortions, and uneven aspect ratios of the features. In the designed MSF decoder, the fusion feature maps of each level of the decoder are generated by fusing the multiscale encoder feature maps and feature map of the next lowest level of the decoder. MSF contributes to the reconstruction of the edge change area by combining feature information at different scales. It was tested on three public VHR remote sensing image CD datasets, and the proposed method demonstrates the best recall and F1-scores, as well as near optimal precision.

As well as very high resolution (VHR) remote sensing technology and deep learning, methods for detecting changes in buildings have made great progress. Despite this, there are still some problems with the incomplete detection of change regions and rough edges. To this end, a change detection network for building VHR remote sensing images based on Siamese EfficientNet B4-MANet (Siam-EMNet) is proposed. First, a bi-branches pretrained EfficientNet B4 encoder structure is constructed to enhance the performance of feature extraction and the rich shallow and deep information is obtained; then, the semantic information of the building is input into the MANet decoder integrated by the dual attention mechanism through the skip connection. The position-wise attention block (PAB) and multi-scale fusion attention block (MFAB) capture spatial relationships between pixels in the global view and channel relationships between layers. The integration of dual attention mechanisms ensures that the building contour is fully detected. The proposed method was evaluated on the LEVIR-CD dataset, and its precision, recall, accuracy, and F1-score were 92.00%, 88.51%, 95.71%, and 90.21%, respectively, which represented the best overall performance compared to the BIT, CDNet, DSIFN, L-Unet, P2V-CD, and SNUNet methods. Verification of the efficacy of the suggested approach was then conducted.