Explainable Artificial Intelligence for Machine Learning-Based Photogrammetric Point Cloud Classification

被引:7
作者
Atik, Muhammed Enes [1 ]
Duran, Zaide [1 ]
Seker, Dursun Zafer [1 ]
机构
[1] Istanbul Tech Univ, Fac Civil Engn, Dept Geomat Engn, TR-34469 Istanbul, Turkiye
关键词
Point cloud compression; Three-dimensional displays; Feature extraction; Deep learning; Buildings; Solid modeling; Explainable AI; Classification; explainable artificial intelligence (XAI); feature selection; machine learning; photogrammetry; point cloud;
D O I
10.1109/JSTARS.2024.3370159
中图分类号
TM [电工技术]; TN [电子技术、通信技术];
学科分类号
0808 ; 0809 ;
摘要
Point clouds are one of the most widely used data sources for spatial modeling. Artificial intelligence approaches have become an important tool for understanding and extracting semantic information of point clouds. In particular, the explainability of machine learning approaches for 3-D data has not been sufficiently investigated. Moreover, existing studies are generally limited to object classification issues. This is a pioneer study that addresses the classification of photogrammetric point clouds in terms of explainable artificial intelligence. In this study, the explainability of black-box machine learning models in the context of the classification of photogrammetric point clouds was investigated. Each point in the point cloud is defined using geometric and spectral features. In addition, the effect of selecting the most important of these features on the classification performance of ML models such as Random Forest, XGBoost, and LightGBM was examined. The explainability of ML models was analyzed with Shapley additive explanation (SHAP), an explainable artificial intelligence approach. SHAP analysis was compared with filter-based information gain (IG) and ReliefF methods for feature selection. Using the features selected with SHAP analysis, overall accuracy (OA) of 85.50% in the Ankeny dataset, 91.70% in the Building dataset, and 83.28% in the Cadastre dataset was achieved with LightGBM. The evaluation with XGBoost shows an OA of 85.22% for Ankeny, 91.21% for Building, and 82.47% for Cadastre. The evaluation with RF shows an OA of 83.70% for Ankeny, 89.08% for Building, and 79.36% for Cadastre.
引用
收藏
页码:5834 / 5846
页数:13
相关论文
共 50 条
[1]   Peeking Inside the Black-Box: A Survey on Explainable Artificial Intelligence (XAI) [J].
Adadi, Amina ;
Berrada, Mohammed .
IEEE ACCESS, 2018, 6 :52138-52160
[2]  
Akagic A., 2022, PROC 28 INT C INF CO, P1
[3]   Low-Cost Laser Scanning System Design [J].
Akyol, Onur ;
Duran, Zaide .
JOURNAL OF RUSSIAN LASER RESEARCH, 2014, 35 (03) :244-251
[4]  
[Anonymous], 2023, 2023 IEEE/CVF Winter Conference on Applications of Computer Vision (WACV), P4229, DOI 10.1109/WACV56688.2023.00422
[5]   Selection of Relevant Geometric Features Using Filter-Based Algorithms for Point Cloud Semantic Segmentation [J].
Atik, Muhammed Enes ;
Duran, Zaide .
ELECTRONICS, 2022, 11 (20)
[6]   An Efficient Ensemble Deep Learning Approach for Semantic Point Cloud Segmentation Based on 3D Geometric Features and Range Images [J].
Atik, Muhammed Enes ;
Duran, Zaide .
SENSORS, 2022, 22 (16)
[7]  
Atik ME, 2021, FRESEN ENVIRON BULL, V30, P4270
[8]   Machine Learning-Based Supervised Classification of Point Clouds Using Multiscale Geometric Features [J].
Atik, Muhammed Enes ;
Duran, Zaide ;
Seker, Dursun Zafer .
ISPRS INTERNATIONAL JOURNAL OF GEO-INFORMATION, 2021, 10 (03)
[9]   Integrating Convolutional Neural Network and Multiresolution Segmentation for Land Cover and Land Use Mapping Using Satellite Imagery [J].
Atik, Saziye Ozge ;
Ipbuker, Cengizhan .
APPLIED SCIENCES-BASEL, 2021, 11 (12)
[10]  
Becker C., 2018, Photogrammetric Eng. Remote Sens., V84, P3