Статті КІТС (ДМетІ)

Permanent URI for this collectionhttp://crust.ust.edu.ua/handle/123456789/14595

Browse

Now showing 1 - 2 of 2

Aircraft Detection with Deep Neural Networks and Contour-Based Methods
(National University "Zaporizhzhia Polytechnic", Zaporizhzhia, 2024) Radionov, Y. D.; Kashtan, Vita Yu.; Hnatushenko, Volodymyr V.; Kazymyrenko, O. V.
ENG: Context. Aircraft detection is an essential task in the military, as fast and accurate aircraft identification allows for timely response to potential threats, effective airspace control, and national security. The use of deep neural networks improves the accuracy of aircraft recognition, which is essential for modern defense and airspace monitoring needs. Objective. The work aims to improve the accuracy of aircraft recognition in high-resolution optical satellite imagery by using deep neural networks and a method of sequential boundary traversal to detect object contours. Method. A method for improving the accuracy of aircraft detection on high-resolution satellite images is proposed. The first stage involves collecting data from the HRPlanesv2 dataset containing high-precision satellite images with aircraft annotations. The second stage consists of preprocessing the images using a sequential boundary detection method to detect object contours. In the third stage, training data is created by integrating the obtained contours with the original HRPlanesv2 images. In the fourth stage, the YOLOv8m object detection model is trained separately on the original HRPlanesv2 dataset and the dataset with the applied preprocessing, which allows the evaluation of the impact of additional processed features on the model performance. Results. Software that implements the proposed method was developed. Testing was conducted on the primary data before preprocessing and the data after its application. The results confirmed the superiority of the proposed method over classical approaches, providing higher aircraft recognition accuracy. The mAP50 index reached 0.994, and the mAP50-95 index reached 0.864, 1% and 4.8% higher than the standard approach. Conclusions. The experiments confirm the effectiveness of the proposed method of aircraft detection using deep neural networks and the process of sequential boundary traversal to detect object contours. The results indicate this approach’s high accuracy and efficiency, which allows us to recommend it for use in research related to aircraft recognition in high-resolution images. Further research could focus on improving image preprocessing methods and developing object recognition technologies in machine learning.
A Neural Network Approach to Semantic Segmentation of Vehicles in Very High Resolution Images
(National University "Zaporizhzhia Polytechnic", Zaporizhzhia, 2025) Kashtan, Vita Yu.; Hnatushenko, Volodymyr V.; Udovyk, Iryna M.; Kazymyrenko, O. V.; Radionov, Y. D.
ENG: Context. The semantic segmentation of vehicles in very high resolution aerial images is essential in developing intelligent transportation systems. It allows for the automation of real-time traffic management and the detection of congestion and emergencies. Objective. This work aims to develop and evaluate the effectiveness of a neural network approach to semantic segmentation in very high resolution aerial images, which provides high detail and correct reproduction of object boundaries. Method. The DeepLab architecture with ResNet-101 as a backbone is used for gradient preservation and multiscale feature analysis. We trained on DOTA data and retrained on specialized sets with classes: vehicles, green areas, buildings, and roads. A loss function based on the Dice coefficient was applied to reduce the imbalance of classes. It effectively solves the class imbalance problem and improves the accuracy of segmenting objects of different sizes. Using ResNet-101 instead of Xception in the backbone network allows us to maintain the gradient as the network depth increases. Results. Experimental studies have confirmed the effectiveness of the proposed approach, which achieves a segmentation accuracy of more than 90%, outperforming existing analogs. The use of multiscale feature analysis allows for preserving the texture features of objects, reducing false classifications. A comparative study with U-Net, SegNet, FCN8s, and other methods confirms the higher performance of the proposed approach in terms of mIoU (82.3%) and Pixel Accuracy (95.1%). Conclusions. The experiments confirm the effectiveness of the proposed method of semantic segmentation of vehicles in ultrahigh spatial resolution images. Using DeepLab v3+ResNet-101 significantly improves the quality of vehicle segmentation in an urbanized environment. Excellent metric performance makes it promising for infrastructure monitoring and traffic planning tasks. Further research will focus on adapting the model to new datasets.