A Novel Framework for Classifying Remote Sensing Images using Convolutional Neural Networks

The classification of Hyperspectral Remote Sensing Images (HRSIs) using Convolutional Neural Networks (CNNs) has gained significant. Due to excessive dimensional data and a lack of sample training examples, the Hughes phenomenon is frequently seen in hyperspectral remote sensing photos. In the realm of high-dimensional information processing, it is worth noting that there is a notable consumption of time and computing power. Additionally, there is a possibility that the features extracted may not accurately represent the data, leading to suboptimal classification efficiency and accuracy. In this research, a novel framework called GCNN is proposed, which combines the concepts of global context spatial attention (GCSA) and three-dimensional convolutional neural network (3DCNN). The GCNN framework aims to enhance the performance of spatial attention and convolutional neural networks by leveraging global context information. By integrating GCSA and 3DCNN, GCNN offers a promising approach for various applications in computer vision and image analysis. The mix-up operation is a commonly employed technique in the field of remote sensing image analysis. Its primary purpose is the spatially heterogeneous data available in these photos, and improve it. This is achieved by transforming the discrete sample space into a continuous one, thereby promoting improved flatness in the local vicinity of the data space. The integration of GCSA into the network allows for the encoding of contextual information from remote sensing scene images into the local features. The findings of this study indicate that the utilization of CNN for remote sensing image recognition yields significantly higher accuracy compared to traditional image recognition models. The recognition accuracy achieved by CNN in this context is reported to be as high as 97%.