Self-Attention and Convolution Fusion Network for Land Cover Change Detection over a New Data Set in Wenzhou, China

Abstract

With the process of increasing urbanization, there is great significance in obtaining urban change information by applying land cover change detection techniques. However, these existing methods still struggle to achieve convincing performances and are insufficient for practical applications. In this paper, we constructed a new data set, named Wenzhou data set, aiming to detect the land cover changes of Wenzhou City and thus update the urban expanding geographic data. Based on this data set, we provide a new self-attention and convolution fusion network (SCFNet) for the land cover change detection of the Wenzhou data set. The SCFNet is composed of three modules, including backbone (local–global pyramid feature extractor in SLGPNet), self-attention and convolution fusion module (SCFM), and residual refinement module (RRM). The SCFM combines the self-attention mechanism with convolutional layers to acquire a better feature representation. Furthermore, RRM exploits dilated convolutions with different dilation rates to refine more accurate and complete predictions over changed areas. In addition, to explore the performance of existing computational intelligence techniques in application scenarios, we selected six classical and advanced deep learning-based methods for systematic testing and comparison. The extensive experiments on the Wenzhou and Guangzhou data sets demonstrated that our SCFNet obviously outperforms other existing methods. On the Wenzhou data set, the precision, recall and F1-score of our SCFNet are all better than 85%.

The Use of Artificial Intelligence and Satellite Remote Sensing in Land Cover Change Detection: Review and Perspectives

“…To address the problem of inconsistent feature scales in LCCD, Yiqun Zhu proposed a fusion model based on self-attention networks and CNN, which can effectively capture multi-scale and local features to refine the results of LCCD [88]. Zhiyong Lv proposed a multi-scale network of LCCD guided by change gradient images, embedding multiscale information attention modules into the U-Net backbone network to merge multi-scale information from dual-date images [89].…”

Multi-Scale Discrete Cosine Transform Network for Building Change Detection in Very-High-Resolution Remote Sensing Images

“…In this context, building change detection (BCD) has received increasing attention in the past decade and can realize effective change monitoring for single-target buildings, including new buildings and disappearing buildings. Therefore, BCD has potential value in many practical applications, such as urban development planning [6,7], evaluation of the urbanization process [8], and urban disaster prevention and mitigation [9,10].…”