Application And Analysis Based On The Combination Of Multi-classifier And PCANet Network

Machine learning has always been one of the hotspots of scholars at home and abroad in recent years,and image recognition and classification is an important research direction in machine learning.In recent years,with offshore oil exploitation and transportation,oil spill accidents occur frequently,and marine disasters are serious.SAR images are characterized by all-weather,all-weather,and strong microwave penetration,and are not affected by climate and light.Oil monitoring;based on machine learning,it has important research significance for oil spill classification of SAR images.In this paper,the shallow learning classifier in machine learning is studied,and combined with PCANet,the SAR image oil spill detection experiment is carried out on the combined model.The main research contents of the article are as follows:(1)A comparative study of common neural network algorithms.In this paper,three typical neural networks,BP neural network,convolutional neural network and wavelet neural network,are selected,and their structural models and algorithm principles are analyzed.The model construction includes two processes,namely forward propagation of signals and back propagation of errors.All three neural networks can learn to extract and classify images,and their network structure is simple or complex,which requires multiple iterative adjustments,weight and threshold.(2)This paper implements four shallow classifiers to detect oil spills on SAR images.The algorithm principles of four classifiers such as Decision Tree,Support Vector Machine,K-Nearest Neighbor and AdaBoost in Ensemble Learning algorithm are analyzed,as well as the decision rules in the image classification process.The classification results of SAR images of HH polarization,HV polarization and VV polarization under three polarization modes are compared and analyzed,and the classification results of different classifiers under the same polarization are analyzed.The results show that the oil spill detection area of the four types of SAR images under different polarizations is larger than the expert interpretation area,and the oil spill effect under the same polarization VV polarization is better than that under the other two polarizations SAR image.(3)This paper implements SAR image oil spill recognition based on PCANet network and shallow classifier.Based on the analysis of the principle and structure of PCANet network algorithm,it is applied to the feature extraction of SAR images under different polarizations,and the extracted feature values are input into different classifiers for SAR image classification.The algorithm is effective.It improves the feature dimension of the data,but its detection accuracy is a bit low,and it is more sensitive to SAR image data under different polarizations.