Remote Sensing Image Target Detection of Polarization Filtering and Cross-dimensional Interactive Shuffling
The multi-channel features extracted by the deep network have weak relevance,weak expression ability,and low accuracy of ship target recognition.To solve the above problems,a remote sensing target detection method based on cross-dimensional interactive shuffling features is proposed in this paper.First,in the feature extraction stage,the equivariant neural network is used to extract features with rotational equivariance,which effectively re-duces the loss of positioning accuracy caused by arbitrary target direction angles.On this basis,the dimensionality collapse method of the polarization filtering mechanism is used to perform weight learning of the extracted features in only space and channel dimensions,which effectively avoids feature loss caused by dimensionality reduction.To establish the mapping relationship between the dimensions,the cross-dimensional interactive operations are used to mine the associated information between the channel and the spatial dimension,and the information between the channels is shuffled before fusion.And as a result,the ability to express features is enhanced and the location ac-curacy of the target is improved.Experiments on the HRSC2016 public data set prove that the detection accuracy of this method is higher than that of the comparison method,which shows the effectiveness of the proposed method.
万方数据
维普
