In the process of the UAV (Unmanned Aerial Vehicle) remote sensing image geometric correction, the method of the geometric correction plays a vital role. Since the neural network is a distributed and parallel mathematical model, and it has good learning ability for nonlinear, so the nonlinear and uncertainty of the UVA remote sensing image geometric correction can be solved well. This paper focuses on the application of BP neural network and RBF neural network in UAV remote sensing image geometric correction, and finally compares the effect of the geometric correction based on BP neural network and RBF neural network through the experiments.

Abstract. This paper focuses on the preprocessing method of control points in geometric correction for UAV (Unmanned Aerial Vehicle) remote sensing image. Control points preprocessing refers to find out the calibration points from the selected control points, so as to use the calibration points to fit the geometric correction function. Because in traditional K-means algorithm, clustering results have a strong dependence on initial clustering centers, so selecting initial clustering centers randomly will lead to the clustering results instability when preprocessing control points with traditional K-means algorithm, and it will influence the effect of the UAV remote sensing image geometric correction. Therefore, the paper imports the thought of Huffman tree to traditional K-means algorithm, aiming at optimizing the selection of initial clustering centers and improving the effect of UAV remote sensing image geometric correction ultimately.

Low Density Parity Check(LDPC) code itself has a good performance and the application of LDPC in the communication field is extensive. LDPC codes have the characteristics of low bit error rate, easy adjustment, low decoding complexity and excellent decoding performance. LDPC is considered to be the "best performance" in the field of coding and decoding [1]. In this paper, we mainly introduce the theory of LDPC codes. And the use of the method of the LDPC code to do a further explanation. Several simple decoding schemes of LDPC are studied. Finally, using the method of function approximation to replace the confidence function to improve the BP LLR algorithm, and simulation experiments are carried out.

Principal component analysis (PCA) is one of the most widely used face feature extraction methods, and has evolved a lot of new algorithms, which has become a hot research topic. It [1] is a multivariate statistical method, can effectively reduce the dimension of the face image, and can keep the original data of most of the major information, has been widely used in the field of pattern recognition and computer vision. This paper introduces the basic principles of PCA, as well as the improved PCA algorithm, and finally the simulation experiments are carried out.