Object detection and classification are among the most popular topics in Photogrammetry and Remote Sensing studies. With technological developments, a large number of high-resolution satellite images have been obtained and it has become possible to distinguish many different objects. Despite all these developments, the need for human intervention in object detection and classification is seen as one of the major problems. Machine learning has been used as a priority option to this day to reduce this need. Although success has been achieved with this method, human intervention is still needed. Deep learning provides a great convenience by eliminating this problem. Deep learning methods carry out the learning process on raw data unlike traditional machine learning methods. Although deep learning has a long history, the main reasons for its increased popularity in recent years are; the availability of sufficient data for the training process and the availability of hardware to process the data. In this study, a performance comparison was made between two different convolutional neural network architectures (SegNet and Fully Convolutional Networks (FCN)) which are used for object segmentation and classification on images. These two different models were trained using the same training dataset and their performances have been evaluated using the same test dataset. The results show that, for building segmentation, there is not much significant difference between these two architectures in terms of accuracy, but FCN architecture is more successful than SegNet by 1%. However, this situation may vary according to the dataset used during the training of the system.

Building segmentation is crucial for applications extending from map production to urban planning. Nowadays, it is still a challenge due to CNNs’ inability to model global context and Transformers’ high memory need. In this study, 10 CNN and Transformer models were generated, and comparisons were realized. Alongside our proposed Residual-Inception U-Net (RIU-Net), U-Net, Residual U-Net, and Attention Residual U-Net, four CNN architectures (Inception, Inception-ResNet, Xception, and MobileNet) were implemented as encoders to U-Net-based models. Lastly, two Transformer-based approaches (Trans U-Net and Swin U-Net) were also used. Massachusetts Buildings Dataset and Inria Aerial Image Labeling Dataset were used for training and evaluation. On Inria dataset, RIU-Net achieved the highest IoU score, F1 score, and test accuracy, with 0.6736, 0.7868, and 92.23%, respectively. On Massachusetts Small dataset, Attention Residual U-Net achieved the highest IoU and F1 scores, with 0.6218 and 0.7606, and Trans U-Net reached the highest test accuracy, with 94.26%. On Massachusetts Large dataset, Residual U-Net accomplished the highest IoU and F1 scores, with 0.6165 and 0.7565, and Attention Residual U-Net attained the highest test accuracy, with 93.81%. The results showed that RIU-Net was significantly successful on Inria dataset. On Massachusetts datasets, Residual U-Net, Attention Residual U-Net, and Trans U-Net provided successful results.

Deep learning architectures are widely used for road segmentation studies. Data sets reflect the characteristics of the study region in which they are generated. The models that are trained with the data sets are unable to accurately forecast the region outside the area covered by the data sets. That is why local data sets must be generated to meet the requirements of navigation, military activities, urban planning and health management applications where road data are indispensable. In this study, satellite images and corresponding mask images of Istanbul city at different zoom levels were generated via Google Map Platform. Trained models were produced using Deep Residual U-Net. This study consists of three parts. The first part includes training each zoom level, from 14 to 17, separately. The recall statistics of models were found as 95%, 94%, 92% and 86%, and corresponding F1-Score statistics were calculated as 95%, 95%, 92% and 86%, respectively. Next, consecutive training with priory weight and $L_0$ regularization on satellite images were tested to increase the prediction results. The consecutive model and $L_0$ regularization methods resulted in an increase in both recall and F1-Score by 2−3%. In the third part, a new model was generated by combining all images which were produced at different levels. The combined model was found successful for the prediction of roads in all zoom levels. Finally, the performance of these models trained with the Istanbul data set was compared with well-known DeepGlobe and Massachusetts road data sets. Hence, this study illustrated that generating all data sets specifically for the area of interest is crucial.

Object recognition is important in many practical applications of computer vision. Traditional 2D methods are negatively affected by illumination, shadowing and viewpoint. 3D methods have the potential to solve these problems, because 3D models include geometric properties of the objects. In this paper, 3D local feature based algorithms were used for 3D object recognition. The local feature was keypoint. This study aimed to research facilities of keypoints for 3D object recognition. Keypoint is feature of object that is detected by detector algorithms according to certain mathematical base. A recognition system was designed. For this purpose, a database that includes 3D model of objects was created. The algorithms were improved in MATLAB. The keypoints on the 3D models were detected using keypoint detectors. These keypoints were described by keypoints descriptors. The descriptor algorithms detect geometrical relation between each point of point cloud and create a histogram. In the third step, the keypoints in different point clouds are matched using the feature histograms obtained. Statistical methods are used to compare generated histograms. Thus, the two closest similar points between the different point clouds are matched. It is expected that the models with the most corresponding points belong to the same object. Euclidean distance between corresponding keypoints in the two point cloud is calculated. It has been accepted that the points are shorter than 10 mm. The positional accuracy of the matched points has been examined. Iterative Closest Point (ICP) was applied to the matching point clouds for this purpose. As a result, the graphics were generated that showed correct matching ratio and root mean square error. As a result, there are different approaches about 3D object recognition in literature. This study aimed to compare different keypoint detector and descriptor algorithms. Intrinsic Shape Signature (ISS) is keypoint detector algorithms. Point Feature Histograms (PFH) and Fast Point Feature Histograms (FPFH) are keypoint descriptor algorithms. The results of this study will provide guidance for future studies.