13 September 2023 in IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing
The classification of land cover types is an important task for monitoring land use. Moreover, with the continuous application of high resolution remote sensing images, the time and space span is becoming larger and larger, which greatly increases the difficulty of classification of target types. And there are few results that can effectively deal with deep land category information. Furthermore, the extraction and fusion of deep features still need to be improved. In this paper, a multi-angle attention fusion network (MAFNet) is proposed for the classification of land cover types. The network uses a 50-layer residual network as a feature extraction network, and an adaptive special-shaped window attention module (ASWA) is added to the deep layer of the network to extract deep semantic information, building the connection between global information. In addition, the multi-angle interactive attention fusion module (MIAF) is used to fuse feature maps at different levels, and an interactive attention mechanism is established at different angles. Finally, a new decoder module (De) is proposed for the decoder to adjust the fused feature map again. Through experiments on three datasets, the results show that the method proposed in this paper is more accurate than the previous network for the segmentation results of different categories. It can realize the accurate division of land types in remote sensing images, and has good generalization ability.
9 March 2022 in Mobile Information Systems
Under big data, a large number of features, as well as their complex data types, make traditional feature extraction and knowledge reasoning unable to adapt to new conditions. To solve these problems, this study proposes a museum big data feature extraction method based on a similarity mapping algorithm. Under the museum big data analysis, the museum big data text information is collected through web crawler technology. The web crawler is used to index the content of websites all across the Internet so that the museum websites can appear in search engine results and the collected text information is denoised and smoothed by a Gaussian filter to construct the processed text information set mapping matrix. The semantic similarity is computed according to the text word concept. Based on the calculation results, through word frequency and document probability inverse document frequency weight, the museum big data text information features are extracted. Simulation results show that the proposed method has high accuracy and short extraction time. Through the comparative analysis, it can be realized that this method not only solves the problems existing in traditional methods but also lays a foundation for the analysis of museum massive data.