Improvement on rooftop classification of worldview-3 imagery using object-based image analysis

Roof materials and roof surface conditions are the main factors that contribute to the determination of rooftop rainwater harvesting (RRWH) quality and quantity in an urban area. Therefore, these factors are required to be analyzed effectively. According to the previous study, the types of different roof materials and roof surface conditions that affected by aging and weathering effects, contribute to the spectral response variability of these features. Even though multispectral images have been used to map the roofing types previously, the characterization of roof surface conditions remains limited because of spectral, spatial, and textural limitations. To overcome such challenges, this study attempts to comprehend and improve the remote sensing technology for rooftop classification using the Worldview-3 (WV-3) image and object-based image analysis (OBIA) method. The improvement process involved segmentation, feature selection and classification techniques. A spatio-statistical optimization technique that combines the Taguchi statistical method and a spatial plateau objective function (POF) was presented to improve the segmentation procedures for building footprint extraction. The optimal scale, shape and compactness parameters of multi-resolution segmentation have been determined and the detection accuracy was evaluated based on receiver operating characteristics (ROC). The result shows the area under the ROC curve (AUC) of 0.804 with p < 0.0001 at 95% confidence level. Furthermore, a systematic feature selection approach was proposed in which search algorithms (Ant-Search, Best First-Search and Particle Swamp Optimization (PSO) - Search) performance were evaluated to select the most significant features. The accuracy of each algorithm was evaluated using LibSVM, Bayes network, and Adaboost classifier. The result presents that the Ant-Search algorithm via LibSVM was determined as the best combination with 100% accuracy. The accuracy of classification result using OBIA is insufficient to depend on the segmentation parameters, the selection of features, and the existence of spectrally mixed objects. An analysis of the choice of classification techniques is also required. Therefore, the LiDAR derived data were combined with WV-3 image using different fusion methods such as layer stacking (LS), Gram–Schmidt (GS), and PC spectral sharpening (PCSS). Then, the classifier (support vector machine (SVM) and data mining (DM) algorithm, decision tree (DT) were applied on each fusion image and their accuracy were evaluated. Generally, the generated DT classification presents a higher overall accuracy with 87%, 72%, and 66% for LS, GS, and PC Pan-Sharpening (PCSS), respectively. Meanwhile, the DT classification using the LS approach produced the highest overall accuracy of 87% and a kappa coefficient of 0.80. Overall, this study offers new insights into remote sensing urban applications, specifically for roof-based mapping through the development of systematic improvement approach using OBIA. Interestingly, the degradation status of the roof in heterogenous urban environments can be determined and the quality of roof-based harvested rainwater affected by roofing materials and roofing conditions can be analyzed effectively.

Text FK 2019 100 - ir.pdf Download (1MB)

Actions (login required)

View Item