An novel random forests and its application to the classification of mangroves remote sensing image

The novel random forests algorithm with variables random input and random combination (Forest_RI_RC) machine was proposed to improve the weakness of low accuracy and over-fitting phenomenon in single decision tree. The proposed method produces more and more selections and combinations to increase the possibility of the best decision-making features. This way reduces the correlation coefficient of the random forests, which efficiently lead to the lower generalization error and approach the higher classification accuracy. The standard machine learning datasets were used to verify the validity of the classification. The simulation results showed that the novel algorithm with the multiple classifiers to concurrently segment the objects and achieve the smaller generalization error. Finally, the algorithm was applied to the classified problems of mangrove remote sensing image. Software simulations presents that the classification accuracy is basically stable at around 90 %. This performance is better than the other two decision tree and bagging methods.