Remote Sensing Scene Data Generation Using Element Geometric Transformation and GAN-Based Texture Synthesis

Classification of remote sensing scene image (RSSI) has been broadly applied and attracted increasing attention. However, classification methods based on convolutional neural networks (CNNs) require a large number manually labeled samples as training data, which is time-consuming costly. Therefore, generating data becomes practical approach. conventional generation generative adversarial (GANs) involve some significant limitations, such distortion limited size. To solve the mentioned problems, herein, we propose method RSSI using element geometric transformation GAN-based texture synthesis. Firstly, segment RSSI, extracting information in RSSI. Then, perform transformations elements extract them. After that, use to model generate texture. Finally, fuse transformed with generated obtain The increases complexity scene. synthesis ensures not distorted. Experimental results demonstrate that by our achieved better visual effect than GAN model. In addition, performance CNN classifiers was reduced 0.44–3.41% enhanced set, partly attributed samples. proposed able diverse sufficient fidelity under conditions small sample size accuracy saturation issues public sets.