DCNR: deep cube CNN with random forest for hyperspectral image classification

摘要: Hyperspectral Image (HSI) classification is one of the fundamental tasks in field remote sensing data analysis. CNN (Convolutional Neural Network) has been proven to be an effective deep learning model, which can extract high-level features directly from raw and thereby utilize rich information contained HSI data. However, labor cost label enough HIS for training model usually expensive, so that it a strong demand utilizing limited get satisfied accuracy. In this paper, we put forward cube – DCNR, composed neighbor pixels strategy, random forest classifier. DCNR cubic samples, containing spectral-spatial information, are generated by putting each target pixel its neighbors together. Then with high representative ability, extracted applying specially designed on sample, fed into classifier pixel. Results show achieve accuracy 96.78%, 96.08% 94.85% KSC, IP SA datasets respectively 20% samples as set, 85.03%, 83.45 62.17% only 1% significantly outperforming models.