Akar, Özlem; Akar, Alper; Bayata, Halim Ferit

This study investigates the use of vegetation indices for accurately identifying crops with similar spectral characteristics as grapes, apricots, tomatoes, wheat, and clover for enhancing crop monitoring and management. A 59-hectare area in Karakaya Village, located in the Üzümlü district of Erzincan Province, Türkiye, was selected as the study area. This area contains crops with varying textures, object height, and spectral characteristics. In the study, multispectral (MS) images were acquired using the SenseFly eBee SQ unmanned aerial vehicle (UAV), and subsequently processed to generate an orthophoto, digital terrain model (DTM), and digital surface model (DSM). Fifteen vegetation indices, Gabor texture features, and object heights were integrated into MS bands. Crop classification was performed using two high-accuracy machine learning algorithms: Random Forest (RF) and Support Vector Machine (SVM). According to the overall classification accuracy results, the use of vegetation indices improved classification accuracy by 9% for RF and 5% for SVM. Incorporating Gabor texture features with the top-performing indices (MACARI1, OSAVI, ADVI, and DVI) further increased accuracy to 20% for RF and 12% for SVM. Additionally, including object height alongside the indices and Gabor features resulted in further accuracy gains of 10% and 11% for RF and SVM, respectively. F1-score, specificity, and accuracy analyses, along with various kappa statistics, also the significant improvements in classification performance. According to the McNemar test, the χ^2 values comparing orthophoto images with those incorporating indices, texture, and object height ranged from 6.353 to 35.556 for RF, and from 7.220 to 11.021 for SVM. Since all χ^2 values exceeded 3.84, the results indicate statistically significant improvements in the classification accuracy at the 95% confidence interval.

Karsli, Fevzi; Ozdemir, Emirhan; Bahadir, Murat

The increasing use of Light Detection and Ranging (LiDAR) point clouds in vector map production highlights the need for a generalized approach to represent complex (linear and curvilinear) building boundary geometries, considering both technological advancements and architectural diversity. In this paper, the aim is to automatically regularize the complex geometric building boundaries extracted from LiDAR point clouds. In order to achieve this aim, building corner points were first detected with the Automatic Building Outline Regularization (ABORE) method. Then, a distance-based statistical method was implemented to determine the edge types representing the building boundary. If the building edges are linear, ABORE corner points are used. Conversely, if they are curvilinear, the edges are regularized using Bézier curves based on Bernstein polynomials. Thus, by generalizing the proposed ABORE method for regularizing only linear building boundaries, a novel method called Generalized Automatic Building Outline Regularization (G-ABORE) is developed to regularize building boundaries in complex geometry. The proposed method was implemented on the open-source Toronto, São Paulo, Minnesota, and locally produced Ordu LiDAR datasets. Regarding accuracy assessments, the overall performance of the novel method for building boundary regularization is 98.76%, 98.87%, 98.11%, and 98.99% for Completeness (Cp), Correctness (Cr), Quality (Q), and F-score metrics, respectively. In addition, RMSE and PoLiS positional accuracy values were calculated as 0.27 m and 0.24 m, respectively. The results demonstrate that the G-ABORE method effectively regularizes complex building boundaries with high accuracy. As a result, it is proven that the developed method can regularize building layers with an accuracy suitable for representation in vector maps, regardless of the structure of building boundary geometries extracted from point cloud data in fields such as architectural design, urban planning, and more.

Akar, Özlem; Saralioğlu, Ekrem; Güngör, Oğuz; Bayata, Halim Ferit

The Erzincan (Cimin) grape, which is an endemic product, plays a significant role in the economy of both the region it is cultivated in and the overall country. Therefore, it is crucial to closely monitor and promote this product. The objective of this study was to analyze the spatial distribution of vineyards by utilizing advanced machine learning and deep learning algorithms to classify high-resolution satellite images. A deep learning model based on a 3D Convolutional Neural Network (CNN) was developed for vineyard classification. The proposed model was compared with traditional machine learning algorithms, specifically Support Vector Machine (SVM), Random Forest (RF), and Rotation Forest (ROTF). The accuracy of the classifications was assessed through error matrices, kappa analysis, and McNemar tests. The best overall classification accuracies and kappa values were achieved by the 3D CNN and RF methods, with scores of 86.47% (0.8308) and 70.53% (0.6279) respectively. Notably, when Gabor texture features were incorporated, the accuracy of the RF method increased to 75.94% (0.6364). Nevertheless, the 3D CNN classifier outperformed all others, yielding the highest classification accuracy with an 11% advantage (86.47%). The statistical analysis using McNemar's test confirmed that the χ2 values for all classification outcomes exceeded 3.84 at the 95% confidence interval, indicating a significant enhancement in classification accuracy provided by the 3D CNN classifier. Additionally, the 3D CNN method demonstrated successful classification performance, as evidenced by the minimum-maximum F1-score (0.79-0.97), specificity (0.95-0.99), and accuracy (0.91-0.99) values.