NEURAL NETWORK AUTOMATION OF FILLING THE DATA SET OF AERIAL DIGITAL IMAGES
DOI:
https://doi.org/10.32689/maup.it.2022.2.12Keywords:
artificial neural network, aerial digital images, pattern recognition, classification, autoencoder.Abstract
In recent years, machine learning technologies and methods have made significant progress. The methods of machine learning and artificial intelligence have been successfully used in a wide and expanding range of areas and applications. An important task of modern science and technology should be considered the use of artificial neural networks to identify, recognize and further classify the various objects present in a particular image. One of the advantages of neural networks is that all elements can operate in parallel, thereby significantly increasing the efficiency of the problem, especially in image processing. Currently, there are quite a number of automatic image recognition systems for various applications. Neural networks can serve as a theoretical and practical basis for the development of such systems. An aspect to pay attention to when implementing neural network recognition technology is the database, which is necessary for neural network training. Today, there is a large amount of data obtained from satellites or aircraft. But unmanned aerial vehicles (UAVs) have their own specifics: they fly at low altitudes, are much more subject to vibration and exposure to air masses. That is, the methods of data selection for network training with subsequent recognition from the onboard cameras of the UAV must take into account these features. The purpose of the article is to describe the developed information technology for automation of filling the training data set of aerial photography for neural network recognition. This paper provides a general plan for the development of information technology. Was used a description of the neural network architecture. The main provisions of software implementation were bented. This work also provides testing on real data, the procedure of replenishment of the dataset by testing segmentation and classification of test images, retraining model and testing classification on new images and the analysis of the results.
References
Chen Z., Xu B. and Gao B., “An image-segmentation-based urban DTM generation method using airborne lidar data”, IEEE Journ. Select. Topics Appl. Earth Observ. Remote Sens., vol. 9 (1), pp. 496–506, January 2016.
Automated Object Recognition System based on Convolutional Autoencoder Prystavka P., Cholyshkina O., Dolgikh S., Karpenko D. 2020 10th International Conference on Advanced Computer Information Technologies, ACIT 2020 – Proceedings, 2020, pp. 830–833, 9208945.
Felzenszwalb P. F. and Huttenlocher D. P. “Efficient graph-based image segmentation”, Int. Journ. Comp. Vision, vol. 59 (2), pp. 167–181, September 2004.
Marfil R., Molina-Tanco L., Bandera A., Rodriguez J. A. and Sandoval F., “Pyramid segmentation algorithms revisited”, Pattern Recognition, vol. 39 (8), pp. 1430–1451, August 2006.
Huang X., Bai H. and Li S. “Automatic aerial image segmentation using a modified Chan-Vese algorithm”, Proceedings of the 9th IEEE Conf. on Indust. Electr. Applic, pp. 1091–1094, June 2014.
Приставка П. О., Чолишкіна О. Г. Поліноміальні сплайни в задачі альтернативної навігації за даними аерозйомки. Монографія. – К. : Міжрегіональна Академія управління персоналом, 2022. – 128 с.
Kalal Z., Mikolaiczyk K. and Matas J. “Tracking-learning-detection”, IEEE Trans. Patt. Anal. Machine Intel., vol. 34 (7), pp. 1409–1422, July 2012.
Krizhevsky A., Sutskever I. and Hinton G. E., “Imagenet classification with deep convolutional neural networks”, Proc. Adv. Neural Inf. Proc. Syst. (NIPS), pp. 1097–1105. Lake Tahoe, Nevada, USA, 2012.
Latent Representations of Terrain in Aerial Image Classification Prystavka P., Dolgikh S., Cholyshkina O., Kozachuk O. CEUR Workshop Proceedingsthis link is disabled, 2021, 3013, pp. 86–95.
“Keras: The Python Deep Learning library”, online: https://keras.io/
Gorban A. N., Kegl B., Wunsch D., Zinovyev A. Y. (Eds.), Principal Manifolds for Data Visualisation and Dimension Reduction, Series: Lecture Notes in Computational Science and Engineering 58, Springer, Berlin – Heidelberg – New York, 2007, XXIV, 340 p. 82 illus.
Електронний ресурс. http://applmaths.nau.edu.ua/index.php
Електронний ресурс. https://drive.google.com/file/d/1BAmSRbYUyCnrPYn-jpHI7l_6qsNmc9o6/view
