Research On Image Recognition And Classification In Strong Noise Environment

With the rapid progress of science and technology and the explosive growth of image information,image processing technology has become an important means of data information acquisition.However,due to various complex uncertainties affecting the generation,transmission and storage,the obtained image information has strong noise interference,which makes the existing image recognition and classification methods ineffective.Based on this,this paper will study image recognition and classification in strong noise environment: for small sample images obtained in real environment are processed,large sample data sets are created through off-line data enhancement,which lays the foundation for feature extraction of image recognition and classification.Two-dimensional Kalman filter(2DKF)sequential multi-sensor information fusion method is used as image preprocessing approach to solve the problems of uneven illumination,attitude transformation and Gaussian noise.Two-dimensional principal component analysis(2DPCA)combined with frame theory and convolution neural network(CNN)are used to recognize and classify images respectively and to improve the accuracy of recognition and classification.Specific research contents and contributions are as follows:(1)Aiming at the problem that the image recognition rate is not high because of the unsatisfactory denoising effect of single sensor image,an image preprocessing method based on 2DKF sequential multi-sensor information fusion is proposed.The state space model of two-dimensional linear discrete system is established by using multi-sensor,and the centralized 2DKF fusion method and the sequential 2DKF fusion method are used for two-dimensional image multi-sensor information fusion denoising.Compared with other methods,the simulation results show that the sequential 2DKF fusion algorithm can improve the peak signal-to-noise ratio(PSNR)numerically,improve the image clarity and obtain more effective denoising effect.(2)Aiming at the similar-image features in strong noise background,a modified 2D PCA image recognition and classification method based on the combination of wavelet transform and frame theory is proposed.Wavelet transform is used to enhance image features,and frame theory is used to interpolate the obtained eigenvectors to obtain redundant and complementary information,so as to improve image recognition rate.K-nearest neighbor method is used for image classification.The simulation results show that,compared with the classical 2DPCA method,this method has higher recognition rate and lower algorithm complexity.(3)Aiming at the inconsistent image sample format problem of this paper studies the image recognition and classification method based on CNN.This method is studied on the basis of CNN method and Caffe carrier,and takes the tank image data set as the research object.Through the training and optimization of image recognition and classification model framework and network parameters,tank image recognition and classification can be realized.The simulation results show that,compared with the 2DPCA method,the CNN-based image recognition and classification method has better recognition rate,when the image data is large and the format is not uniform.