Abstract:In order to improve the accuracy of image classification problem, this paper proposes a new classification method based on image feature extraction and neural network. The method consists of two stages: image feature extraction and neural network classification. At the stage of feature extraction, spatial pyramid matching (SPM) feature, local position feature and global contour feature are extracted. The utilization of spatial information in SPM feature is effectively improved by combining the above three features. At the stage of neural network classification, a multi-hidden layer feedforward Histogram Intersection Kernel Weighted Learning Network (HWLN) is proposed to take advantage of three features to improve the classification accuracy. In the structure of network, the hidden layer output features are used as the bias of the input features to weight the coefficients of the features. The input layer and the hidden layers are directly connected with the output layer to realize the combination of linear mapping and nonlinear mapping. And the Histogram Intersection Kernel is used instead of random initialization of the input weight matrix. Taking combined features as input information, HWLN can realize the mapping relationship between input information and target category, so as to complete the image classification task. Extensive experiments are performed on Caltech 101, Caltech 256 and MSRC databases respectively. The experimental results show that the proposed method further utilizes the spatial information, and thus improves the accuracy of image classification.

Research on Recognition Method Based on Spectral Regression and Back Propagation Neural Network

Robust spectral regression for face recognition

Research on Speech Recognition Methods Based on Spectral Subtraction and Improved BP Neural Network Algorithm

Superpixel-Based Long-Range Dependent Network for High-Resolution Remote-Sensing Image Classification

Temporal Coding of Local Spectrogram Features for Robust Sound Recognition

Image Classification Method Rationally Utilizing Spatial Information of the Image

Deep Residual Attention Network For Spectral Image Super-Resolution

SPECTRAL-SPATIAL MULTISCALE RESIDUAL NETWORK FOR HYPERSPECTRAL IMAGE CLASSIFICATION

Face recognition fellow-up classification algorithm based on BP neural network

Pixel-Level Remote Sensing Image Recognition Based on Bidirectional Word Vectors

Spectral Analysis and Recognition Using Multi-scale Features and Neural Networks

Research and Application of Object Recognition Method Based on Depth Neural Network

Spatial Proximity Feature Selection With Residual Spatial-Spectral Attention Network for Hyperspectral Image Classification.

Relationship Learning From Multisource Images via Spatial-Spectral Perception Network

Space Target Recognition Based on Radar Network Systems With BiGRU-Transformer and Dual Graph Fusion Network

Spectral Regression for Efficient Regularized Subspace Learning

BP neural network based on principle component analysis in multi-spectral remote sensing images classification

Adaptive Neighborhood-Preserving Discriminant Projection Method for HRRP-Based Radar Target Recognition

Enhancing performance of the backpropagation algorithm via sparse response regularization.

Hierarchical Recognition System for Target Recognition from Sparse Representations

Improvements and applications of BP neural network algorithm in radar target recognition