Abstract Buildings that are constructed without the necessary permits and building inspections affect many areas, including safety, health, the environment, social order, and the economy. For this reason, it is essential to determine the number of buildings and their boundaries. Determining the boundaries of a building based solely on its location in the world is a challenging task. In the context of this research, a new approach, BBD, is proposed to detect architectural objects from large-scale satellite imagery, which is an application of remote sensing, together with the geolocations of buildings and their boundaries on the Earth. In the proposed BBD method, open-source GeoServer and TileCache software process huge volumes of satellite imagery that cannot be analyzed with classical data processing techniques using deep learning models. In the proposed BBD method, YOLOv5, DETR, and YOLO-NAS models were used for building detection. SAM was used for the segmentation process in the BBD technique. In addition, the performance of the RefineNet model was investigated, as it performs direct building segmentation, unlike the aforementioned methods. The YOLOV5, DETR and YOLO-NAS models in BBD for building detection obtained an f1 score of 0.744, 0.615, and 0.869 respectively on the images generated by the classic TileCache. However, the RefineNet model, which uses the data generated by the classic TileCache, achieved an f1 score of 0.826 in the building segmentation process. Since the images produced by the classic TileCache are divided into too many parts, the buildings cannot be found as a whole in the images. To overcome these problems, a fine-tuning based optimization was performed. Thanks to the proposed fine-tuning, the modified YOLOv5, DETR, YOLO-NAS, and RefineNet models achieved F1 scores of 0.883, 0.772, 0.975 and 0.932, respectively. In the proposed BBD approach, the modified YOLO-NAS approach was the approach that detected the highest number of objects with an F1 score of 0.975. The YOLO-NAS-SAM approach detected the boundaries of the buildings with high performance by obtaining an IoU value of 0.912.

Summary Object detection is a type of application that includes computer vision and image processing technologies, which deal with detecting, tracking, and classifying desired objects in images. Computer vision is a field of artificial intelligence that enables computers and systems to derive information from digital images and take action or suggestions based on that information. CNN is one of the current methods of object detection due to its ease of use and GPU‐supported parallel working features. Due to the aim of completing deep learning model training quickly or due to insufficient dataset, many studies using the transfer learning method are carried out in fields such as medicine, agriculture, and weapons. However, there are very few studies that use the fine‐tuning method and compare transfer learning in terms of effectiveness. By paying attention to the balanced distribution of the data, approximately 100 images of each chess piece type were included in the analysis and a dataset of at least 1000 images was created. The without transfer learning fine‐tune, fine‐tuned transfer learning, transfer learning, fully supervised learning (FSL) and weakly supervised learning (WSL) applied models performances compared. Experimental results show that the fine‐tuned transfer learning applied YOLO V4 model produces more accurate results than the other models in FSL and the transfer learning applied Faster R‐CNN model produces more accurate results than the other models in WSL.

Summary Android malware has become a serious threat to mobile device users, and effective detection and defence architectures are needed to solve this problem. Recently, machine learning techniques have been widely used to deal with Android malicious apps. These methods are based on a simple feature set and have difficulty detecting up‐to‐date malware. Therefore, more robust and efficient classification methodologies are needed. In this article, AMD‐CNN, an Android malware detection tool, is proposed, and it uses graphical representations to detect malicious apks. In the first step, the features related to the androidmanifest.xml file are extracted and converted into a vector consisting of one or zero. The feature vector is then converted to 2D‐code images and used in training the CNN network. The model needs low‐resource consumption to run on mobile devices and allow real‐time applications to be analyzed. The experiments with 1920 malicious and benign apks show that the malware detection rate (accuracy) was 96.2% and precision, recall, and F‐score values were 97.9%, 98.2%, and 98.1%, respectively. The average time and memory space to analyze each application are 0.035 s and 3.38 MB. AMD‐CNN is an efficient and robust tool and has advantages over previous studies.