Application of the deep learning method in solving the problem of identifying landmarks in the image

Authors: Nguyen Quy Thanh
Published in issue: #11(88)/2023
DOI: 10.18698/2541-8009-2023-11-950
Category: Informatics, Computer Engineering and Control | Chapter: Information Technology. Computer techologies. Theory of computers and systems
Keywords: computer vision, convolutional neural network, image recognition, object detection, image processing, deep learning, artificial intelligence methods, DELF
Published: 19.12.2023

The paper considers an approach to identifying landmarks in the images using the convolutional neural networks. It presents an overview of the algorithm main phases including data preparation, selecting the CNN architecture, model learning, and results evaluation. Image preprocessing methods such as removing noise and background objects are described. Examples of the popular CNN architectures are provided. The learning phase includes weight initialization, forward and backward propagation and weights updating to minimize losses. To assess the classification accuracy, the following metrics are used: accuracy, precision, and recall. Approaches to improving the results are considered: data increase, network structure alteration, and image quality improvement. Importance of recognizing landmarks in tourism, architecture, and heritage is emphasized. A review analysis of introducing deep learning in automation of this task is presented.

References

[1] Raspoznavanie stsen na izobrazheniyakh s pomoshch’yu glubokikh svertochnykh neyronnykh setey [Scene recognition in images using deep convolutional neural networks]. URL: https://habr.com/ru/companies/vk/articles/359214/ (accessed May 19, 2023).

[2] Noh H., Araujo A., Sim J., Weyand T., Han B. Large-Scale Image Retrieval with Attentive Deep Local Features. ICCV, 2017, pp. 3456–3465.

[3] Weyand T., Araujo A., Cao B., Sim J. Google Landmarks Dataset v2 – A Large-Scale Benchmark for Instance-Level Recognition and Retrieval. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2020, pp. 2575–2584.

[4] Google Landmark Recognition 2021. URL: https://www.kaggle.com/c/landmark-recognition-2021/overview/iccv-2021 (accessed May 19, 2023).

[5] Xu C., Wang W., Liu S., Wang Y., Tang Y., Bian T., Yan Y., She Q., Yang C. 3rd Place Solution to Google Landmark Recognition Competition 2021. arXiv preprint arXiv:2110.02794, 2021.

[6] AlexNet — Svertochnaya neyronnaya set’ dlya klassifikatsii izobrazheniy [AlexNet — Convolutional neural network for image classification]. URL: https://neurohive.io/ru/vidy-nejrosetej/alexnet-svjortochnaja-nejronnaja-set-dlja-raspoznavanija-izobrazhenij/ (accessed May 19, 2023).

[7] VGG Very Deep Convolutional Networks (VGGNet) — what you need to know. URL: https://viso.ai/deep-learning/vgg-very-deep-convolutional-networks/ (accessed May 19, 2023).

[8] ResNet (34, 50, 101): ostatochnye CNN dlya klassifikatsii [ResNet (34, 50, 101): residual CNNs for classification]. URL: https://neurohive.io/ru/vidy-nejrosetej/resnet-34-50-101/ (accessed May 19, 2023).

[9] Nguyen T.C., Syryamkin V.I., Nguyen T.H.T. Model for the method of object recognition in images using convolution neuron network — CNN. Modern high technology, 2020, vol. 12–2, pp. 269–280. (In Russ.). http://doi.org/10.17513/snt.38445

[10] Sharibaev R.N., Juraev Sh.S., Tokhirzhonova M.R. Improving classification by cocoon varieties using convolutionary neural networks. Theory and practice of modern science, 2023, no. 6 (96), pp. 212–214. (In Russ.).