Depth separable convolution hyperspectral image classification method based on residual connection

Abstract

The invention belongs to the technical field of hyperspectral image classification, and discloses a depth separable convolution hyperspectral image classification method based on residual connection,and the method comprises the steps: constructing a classification model; wherein the first layer of the classification model adopts 1*1 convolution and then is connected with a ReLU activation function for extracting nonlinear features of spectral information; adopting three residual error units with pyramid structures, and adopting two 3*3 convolution with separable depth in each residual error unit for extracting spectral-space information in the image; wherein at the tail end of the classification model, 1*1 convolution and a global average pooling layer are combined to be used for fusing spatial-spectral features, and classification is completed; and completing hyperspectral image classification through the constructed classification model. According to the method, the spatial-spectralcharacteristics of the spectral image of the hyperspectral image are fully extracted, a lightweight classification model is constructed by stacking the residual units, and the model has higher classification speed while higher precision is ensured.

Description

technical field [0001] The invention belongs to the technical field of hyperspectral image classification, and in particular relates to a depth-separable convolutional hyperspectral image classification method based on residual connections. Background technique [0002] Hyperspectral imaging technology can detect the two-dimensional geometric space information and one-dimensional continuous spectral information of the target object at the same time, making hyperspectral images have the characteristics of "map-spectrum integration". Geometric space information can reflect the size, shape and other external characteristics of the target object, while spectral information can reflect the internal physical structure and chemical composition of the target object. Therefore, hyperspectral remote sensing is widely used in rock mineral detection, marine plant detection, water resource application and land resource utilization and other fields. [0003] How to construct a more accur...

AI Technical Summary

Problems solved by technology

[0005] The present invention aims at the traditional classification algorithm that often changes the band correlation of the original image, loses part of the spectral information, and cannot fully extract the abstract features of the hyperspectral image, thus affecting the classification accuracy; and the current CNN-based classification model has a large Computational overhead and high requirements for computer hardware equipment, a deep separable convolutional hyperspectral image classification method based on residual connections is proposed

Images