SAR4LCZ-Net: A Complex-Valued Convolutional Neural Network for Local Climate Zones Classification Using Gaofen-3 Quad-Pol SAR Data

Abstract

The recent local climate zones (LCZ) classification scheme provides spatially fine granular descriptions of innerurban morphology. It is universally applicable to cities worldwide and capable of supporting various urban studies. Although optical and dual-pol SAR data continue to push the frontiers of this task, the potential of quad-pol SAR data for LCZ classification is not yet explored. In this paper we propose a novel complex-valued convolutional neural network (CNN), SAR4LCZ-Net, to tackle this challenge. SAR4LCZ-Net improves the state of the art by exploiting two facts of this specific task: the semantic hierarchical structure of the LCZ classification scheme, and the complex-valued nature of quad-pol SAR data. To validate the performance of our algorithm, we generate a Chinese Gaofen-3 quad-pol SAR data set for LCZ which covers 31 cities around the world. Results show that the proposed SAR4LCZ-Net improves 2.4% on overall accuracy and 4.5% on average accuracy compared to the real-valued CNN with same structure. Gaofen-3 quad-pol SAR data also showed its advantage over the dual-pol Sentinel-1 data. It enhanced 5.0% on overall accuracy and 7.2% on average accuracy in LCZ classification, under a fair comparison with a model trained by Sentinel-1 of the same area.Index Terms-Quad-pol SAR, complex-valued convolutional neural networks, local climate zones, urban land cover. I. INTRODUCTIONL OCAL climate zones (LCZ) provides an universally applicable classification system for the urban environment. The 17 LCZ classes are based on climate-relevant surface properties on the local-scale, mainly related to the

A Novel Weighted Amplitude Modulation (WAM) System for Ambiguity Suppression of Spaceborne Hybrid Quad-Pol SAR

“…In hybrid quad-pol SAR the left-and right-circular polarization pulses are alternately emitted, and received by two orthogonal linear polarized antennas (e.g., H-and V-polarized antennas) at the same time after scattering by targets [17]. After that, the obtained full scattering matrixes are processed to obtain polarimetric SAR images, thereby implementing some missions, such as Pol-InSAR [18,34], climate zones classification [12], rice mapping [16], and so on.…”

“…At this point, the echo signals in the linear polarized basis can be derived by multiplying the weighting matrices of Equations ( 11) and ( 12) on the right side of (10), which can be represented by Equations ( 13) and ( 14) respectively. γ i and µ i represent the intermediate variables, and they can be calculated according to ( 10)- (12).…”

“…The results of the first imaging mode are shown in Figure 6. It should be noted that only Figure 6a was measured by SAR satellite [12], which serves as the input to simulations, and the simulations were manipulated by the computer that was configured with i7-7700 CPU, eight cores, eight threads, and 32-GB RAM. Some special areas in Figure 6 are highlighted by the red dotted boxes to evaluate the azimuth ambiguity suppression capability of the proposed SAR system.…”

“…The additional degrees of freedom of its transceivers can provide several advantages, including the enhancement of classification/discrimination between targets and clutters [4][5][6], polarization coding on transmit [7], measurement of scattering depolarization for remote sensing applications [8][9][10][11], and so on. Based on quad-pol SAR data, a lot of applications can be performed, including climate zones classification [12], extraction of surface roughness [13], soil moisture mapping [14], moving target indication [15], and rice mapping [16].…”

Complex-Valued Vs. Real-Valued Convolutional Neural Network for Polsar Data Classification

“…Later, the bottleneck features of the CV-CAE are extracted and used in a one layer fully connected classification network. Moreover, Zhang et al [14] proposed a novel CV-CNN architecture, called "SAR4LCZ-Net", for Local Climate Zones (LCZ) classification in Gaofen-3 quad-pol SAR images. Furthermore, Sun et al [15] proposed a complex-valued generative adversarial network (GAN) and utilized a semisupervised classification procedure for PolSAR data classification.…”