Rice identification in hilly areas based on multi-source remote sensing fea-ture fusion and convolutional neural networks(CNN)
To investigate the effect and applicability of the fusion of convolutional neural networks(CNN)algorithm and multi-source remote sensing preferred feature data on the recognition of rice growing areas in hilly areas,we took Shanggao County,Jiangxi province as the study area,and used Sentinel-2 and GF-1 remote sensing image data to identify the late rice growing areas in the study area.Classification features such as image band features,vegetation indices,texture features and terrain features were se-lected,and the feature variables with greater separation for each category were screened out as the preferred feature set using the seperability and thresholds(SEaTH)algorithm.The fusion of Sentinel-2 and GF-1 preferred features and Sentinel-2 and GF-1 pre-ferred features were compared with the CNN classification algorithm for late rice recognition,and the support vector machine(SVM)and maximum likelihood(MLC)classification algorithms were used to compare the results.The results showed that the fusion data of Sentinel-2 and GF-1 preferred features had the best recognition effect on rice under CNN classification algorithm.The overall accuracy and Kappa coefficient were 96.19%and 0.93,respectively,and the accuracy was 94.69%when combined with the field survey data for validation.According to the research results,the fusion of Sentinel-2 and GF-1 preferred features had good effect and applicability for rice recognition in hilly areas under CNN classification algorithm,and was an effective tool for remote sensing recognition of rice in hilly areas.
万方数据
维普
