Integration of Ant Colony Optimization and Object-Based Analysis for LiDAR Data Classification

ABSTRACT

Light detection and ranging (LiDAR) data classification provides useful thematic maps for numerous geospatial applications. Several methods and algorithms have been proposed recently for LiDAR data classification. Most studies focused on object-based analysis because of its advantages over per-pixelbased methods. However, several issues, such as parameter optimization, attribute selection, and development of transferable rulesets, remain challenging in this topic. This study contributes to LiDAR data classification by developing an approach that integrates ant colony optimization (ACO) and rule-based classification. First, LiDAR-derived digital elevation and digital surface models were integrated with high-resolution orthophotos. Second, the processed raster was segmented with the multiresolution segmentation method. Subsequently, the parameters were optimized with a supervised technique based on fuzzy analysis. A total of 20 attributes were selected based on general knowledge on the study area and LiDAR data; the best subset containing 12 attributes was then selected via ACO. These attributes were utilized to develop rulesets through the use of a decision tree algorithm, and a thematic map was generated for the study area. Results revealed the robustness of the proposed method, which has an overall accuracy of ∼95% and a kappa coefficient of 0.94. The rule-based approach with all attributes and the k nearest neighbor (KNN) classification method were applied to validate the results of the proposed method. The overall accuracy of the rule-based method with all attributes was ∼88% (kappa = 0.82), whereas the KNN method had an overall accuracy of

<70% and produced a poor thematic map. The selection of the ACO algorithm was justified through a comparison with three well-known feature selection methods. On the other hand, the transferability of the developed rules was evaluated by using a second LiDAR dataset at another study area. The overall accuracy and the kappa index for the second study area were 92% and 0.90, respectively. Overall, the findings indicate that the selection of a subset with significant attributes is important for accurate LiDAR data classification with object-based methods.>

INTRODUCTION

Light detection and ranging (LiDAR) data classification provides important information for several environmental studies and a wide range of applications, such as urban mapping [1]–[4], forest classification [5], land suitability analysis [6], and natural hazard and risk assessment [7]. The main product of LiDAR data classification is the landuse/landcover (LULC) map, which is the basic geographic information system (GIS) layer in numerous applications. Although LULC can be extracted from various sources, such as optical satellite images (e.g., multispectral and hyperspectral), synthetic aperture radar images, unmanned aerial vehicle images, and aerial orthophotos, LiDAR data have several advantages in LULC mapping. These advantages include highly accurate digital surface models (DSMs) and high-resolution orthophotos that are minimally affected by shadows and illuminations. Moreover, LiDAR data can be acquired at a user-specified time, and the light source can be easily controlled. Although LiDAR systems have advantages over other systems for LULC mapping, they require robust algorithms compared with other systems because of the lack of spectral information. As a result, several methods and algorithms have been proposed by many researchers to effectively classify LiDAR data and produce accurate LULC maps. Most of the methods proposed recently are generally based on object-based image analysis (OBIA) because of the advantages of OBIA in LiDAR data processing. OBIA has several advantages over per-pixel-based methods. First, OBIA interprets images like the human brain does by including spatial, textural, and contextual information in the feature extraction process [1]. Second, data from different sources can be easily integrated [8]. Third, the final product is in a GIS-ready format and can be utilized for vector- and rasterbased analyses. Fourth, multiscale segmentation allows for the extraction of detailed information from high-resolution images [9]. However, several issues in the OBIA process require optimization and careful analysis; these issues include parameter optimization, attribute selection, and ruleset development. The selection of segmentation parameters directly affects the quality of image segments, and subsequently, the classification results. Adding extra attributes does not necessary improve classification accuracy; only significant attributes must be selected for accurate feature extraction. Nonsignificant attributes are similar to noise data. They result in misclassifications in the classification process and degrade the accuracy of classification. Developing reliable rulesets requires adequate knowledge about the study area. These rulesets should be carefully developed, and their transferability aspect should be considered.

METHODS

Fig. 2 shows the overall workflow of LiDAR data classification with ACO and OBIA. First, LiDAR point clouds are filtered to separate the ground points from the nonground points. After the LiDAR point clouds are classified as ground and nonground, the digital elevation model (DEM), DSM, and height feature (nDSM) are produced by converting LiDAR points into raster data through inverse distance weighting (IDW) interpolation [26]. In this research, the IDW interpolation technique was used owing to its popularity in this field [27]. Afterward, bands are composited to create the processed raster, which is subsequently used for segmentation and classification. Then, the processed raster is segmented with the multiresolution segmentation algorithm (MRS) implemented in eCognition software, and its parameters are optimized with the supervised approach proposed by [28]. After creating image segments, several attributes, including spectral, spatial, textural, and contextual, are calculated. The best subset of attributes is then selected with the ACO method. Significant attributes and image objects are utilized to develop rulesets that define image features using the decision tree (DT) algorithm. The developed rulesets are applied to produce the final LULC map for the study area. Finally, the LULC map is validated with ground-truth data obtained from the field and Google maps.

CONCLUSION

LiDAR classification data provide basic GIS information for numerous geospatial applications, including urban planning, site selection, natural hazard and risk assessment, and LULC change detection. A classification method that integrates ACO and OBIA for LiDAR data classification was developed in this study. ACO was selected for its efficiency and advantage over several other techniques, including GAs, in selecting the best attribute subsets. OBIA was used to classify image objects by incorporating spectral, spatial, textural, and contextual attributes to improve the classification accuracy significantly. In addition, segmentation parameters (scale, shape, and compactness) were optimized with a supervised fuzzy-based method. The rule-based classification method was adopted for OBIA. The experimental results showed that this integrated approach improved the overall accuracy of classification by 7% and was more accurate than the supervised KNN method. The overall accuracy of the integrated approach was 88%, whereas that of the KNN method was 71.90%. In addition, the overall accuracy and the kappa index for the second study area were 92% and 0.90, respectively. In general, the findings indicate that the selection of a subset with significant attributes is important for accurate LiDAR data classification with object-based methods. The limitation of the proposed method is the lack of extracted parking lots with high quality and accuracy; its segmentation performance should also be improved in the future work. Another means to improve LiDAR data classification could be the development of hybrid models that consider the optimization of segmentation parameters, employ multiscale analysis, select the best attribute subset, and design transferable rulesets.