Research on SVM Classification Technology with Improved Shoreland Feature Selection

Aiming at the application requirements of nautical chart revision and marine surveying and mapping, the identification of beach features based on hyperspectral remote sensing images has the advantages of high spatial resolution, high spectral resolution and wide coverage, which has important application prospects in the recognition of shore and beach features. Due to the complexity of coastal features, it is difficult for traditional classification methods to find qualified uniform plots homogeneous parcel for sampling homogeneous parcel sampling on hyperspectral images of coastal zone, resulting in unsatisfactory classification results. SVM classification can reasonably control the generalization ability of the classifier according to the number of samples, but it is sensitive to noise. This paper attempts to process the images with the MNF transform improved by the spatial-spectral dimensional decorrelation method based on noise assessment to reduce the influence of noise on the subsequent classification work, and make the coastal zone shoreling linearly separable, so as to ensure the classification accuracy and improve the classification efficiency.