研究目的
To propose an ensemble-based approach for urban land use and land cover classification that includes traditional classifiers as well as several advanced architectures of Neural Networks.
研究成果
The proposed multi-sensor urban land use and land cover classification approach based on an ensemble of multiple classifiers demonstrated high operational performance and generalization capabilities.
研究不足
Some classes are not well classified due to class imbalance. Major thoroughfares, roads and highways have mutual confusions since the actual class depends on the context only that in many cases is ambiguous.
1：Experimental Design and Method Selection:
The approach includes an ensemble of Gradient Boosting Machines, a set of weak Random Forest classifiers and Convolutional Neural Networks. Various post-processing techniques are applied to improve the classification result.
2：Sample Selection and Data Sources:
The dataset provided by IEEE GRSS 2018 Data Fusion Contest technical committee, covering the University of Houston campus and its surrounding areas, was used.
3：List of Experimental Equipment and Materials:
RGB imagery, raw and processed Multispectral LiDAR data, and Hyperspectral imagery were used.
4：Experimental Procedures and Operational Workflow:
Preprocessing and feature extraction were performed on the data. Cross-validation strategies were utilized to evaluate the models' performance.
5：Data Analysis Methods:
The combination of the models was done on the measurement level using weights for every class of each model obtained from the CV process.
独家科研数据包，助您复现前沿成果，加速创新突破
获取完整内容
