Classification method and system of UAV hyperspectral vegetation species based on deep learning
The application discloses a classification method and a system of UAV hyperspectral vegetation species based on a deep learning, where the method includes the following steps: collecting hyperspectral images by a UAV; preprocessing collected hyperspectral images to obtain preprocessed images, and performing a stitching mosaicking preprocessing on the preprocessed images to obtain hyperspectral orthoimages; labeling the hyperspectral orthoimages to obtain a label data set; performing a vegetation index fusion on the hyperspectral orthoimages to obtain vegetation index-hyperspectral orthoimages; constructing a grassland vegetation classification model based on the vegetation index-hyperspectral orthoimages and the label data set, and completing a vegetation species classification by using the grassland vegetation classification model.
1. A classification method of UAV hyperspectral vegetation species based on a deep learning, comprising following steps:
collecting hyperspectral images of vegetations in a sample area by a UAV and an airborne hyperspectral instrument;
preprocessing collected hyperspectral images to obtain preprocessed images, and performing a stitching mosaicking preprocessing on the preprocessed images to obtain hyperspectral orthoimages;
labeling the hyperspectral orthoimages to obtain a label data set; wherein a method for obtaining the label data set comprises: firstly, importing the hyperspectral orthoimages into a label labeling software, extracting spectral features of the hyperspectral orthoimages, and performing an artificial visual interpretation according to the spectral features of different vegetation species in the label labeling software and labeling the different vegetation species to obtain a region of interest of each of the vegetation species in the images; then, converting a labeled region of interest into a label grid file; finally, after the label grid file is obtained, dividing the label grid file into a training set and a test set;
performing a vegetation index fusion on the hyperspectral orthoimages to obtain vegetation index-hyperspectral orthoimages; wherein a method for obtaining the vegetation index-hyperspectral orthoimages comprises: firstly, selecting infrared and near-infrared bands for the hyperspectral orthoimages, and then calculating a normalized differential vegetation index value, a difference vegetation index value and a ratio vegetation index value for the hyperspectral orthoimage according to the selected infrared and near-infrared bands to obtain corresponding grid images respectively; then, fusing obtained grid images into bands of the hyperspectral orthoimages in a form of band to obtain the vegetation index-hyperspectral orthoimages; and
constructing a grassland vegetation classification model based on the vegetation index-hyperspectral orthoimages and the label data set, and completing a vegetation species classification by using the grassland vegetation classification model; wherein the grassland vegetation classification model comprises a mobile 3D atrous convolution vision Transformer model, wherein the mobile 3D atrous convolution vision Transformer model has a multi-level structure design and comprises three stages and five modules, comprising a 3D convolution, a 3D atrous convolution, a mobile convolution vision Transformer, a reverse residual structure and a convolution vision Transformer; wherein stage 1 comprises the 3D convolution, the 3D atrous convolution and a mobile convolution vision Transformer module, stage 2 comprises the 3D atrous convolution and the mobile convolution vision Transformer module, and stage 3 comprises the reverse residual structure and a convolution vision Transformer module.
2. The classification method of the UAV hyperspectral vegetation species based on the deep learning according to claim 1 , wherein a preprocessing method comprises a lens calibration, a reflectivity calibration and an atmospheric calibration.
3. The classification method of the UAV hyperspectral vegetation species based on the deep learning according to claim 1 , wherein a method for obtaining the hyperspectral orthoimages comprises: importing the preprocessed images into a mosaic software, inputting a flying height of a UAV when obtaining images, and manually eliminating abnormal images; then setting stitching parameters and previewing an effect until stitching results reach an expectation; finally, performing a full-band stitching mosaicking of the images to generate the hyperspectral orthoimages of a research area.
4. A classification system of UAV hyperspectral vegetation species based on a deep learning, used to realize the method of claim 1 , comprising an acquisition module, a stitching module, a labeling module, a fusion module and a construction module;
the acquisition module is used for obtaining hyperspectral images by using the UAV;
the stitching module is used for performing a preprocessing and a stitching mosaicking on the hyperspectral images to obtain hyperspectral orthoimages;
the labeling module is used for labeling the hyperspectral orthoimages to obtain a label data set;
wherein a process of obtaining the label data set comprises: firstly, importing the hyperspectral orthoimages into a label labeling software, extracting spectral features of the hyperspectral orthoimages, and performing an artificial visual interpretation according to the spectral features of different vegetation species in the label labeling software and labeling the different vegetation species to obtain a region of interest of each of the vegetation species in the images; then, converting a labeled region of interest into a label grid file; finally, after the label grid file is obtained, dividing the label grid file into a training set and a test set;
the fusion module is used for performing a vegetation index fusion on the hyperspectral orthoimages to obtain vegetation index-hyperspectral orthoimages; wherein a process of obtaining the vegetation index-hyperspectral orthoimages comprises: firstly, selecting infrared and near-infrared bands for the hyperspectral orthoimages, and then calculating a normalized differential vegetation index value, a difference vegetation index value and a ratio vegetation index value for the hyperspectral orthoimage according to the selected infrared and near-infrared bands to obtain corresponding grid images respectively; then, fusing obtained grid images into bands of the hyperspectral orthoimages in a form of band to obtain the vegetation index-hyperspectral orthoimages; and
the construction module is used for constructing a grassland vegetation classification model based on the vegetation index-hyperspectral orthoimages and the label data set, and completing a vegetation species classification by using the grassland vegetation classification model; wherein the grassland vegetation classification model comprises a mobile 3D atrous convolution vision Transformer model, wherein the mobile 3D atrous convolution vision Transformer model has a multi-level structure design and comprises three stages and five modules, comprising a 3D convolution, a 3D atrous convolution, a mobile convolution vision Transformer, a reverse residual structure and a convolution vision Transformer; wherein stage 1 comprises the 3D convolution, the 3D atrous convolution and a mobile convolution vision Transformer module, stage 2 comprises the 3D atrous convolution and the mobile convolution vision Transformer module, and stage 3 comprises the reverse residual structure and a convolution vision Transformer module.