N.A. Andriyanov1, Yu.B. Kamalova2

1,2 Financial University under the Government of the Russian Federation (Moscow, Russia)

Problem. Currently, not only the quality of computer vision systems, but also their speed is of particular interest. In the task of classifying pollen grains, where the separation of microscopic images is currently performed by operators, the speed of machine processing is very important. The faster the system can classify input pollen grain images, the greater the time savings. It is clear that in this case the quality (accuracy) should not be lower than the specified threshold.Currently, not only the quality of computer vision systems, but also their speed is of particular interest. In the task of classifying pollen grains, where the separation of microscopic images is currently performed by operators, the speed of machine processing is very important. The faster the system can classify input pollen grain images, the greater the time savings. It is clear that in this case the quality (accuracy) should not be lower than the specified threshold.

Target. The aim of this work is to analyze the qualitative and performance characteristics of algorithms based on the convolutional neural network VGG-19 in the recognition of pollen grains.

Results. The article is aimed at analyzing the problem of classifying images of pollen grains obtained from an optical microscope. The study is devoted to the development of efficient and high-speed pollen grain recognition algorithms. The transfer learning method of the VGG-19 neural network is compared with different hyperparameter settings. The use of the VGG-19 neural network ensured the share of correct recognitions at the level of 79.4%. Based on the augmentation of the original samples, it was possible to increase the recognition accuracy by 3.1%. In this case, the optimization is performed by quantizing the weight coefficients of the neural network. Losses in accuracy amounted to 1.1%, but the processing speed increased by about 2.15 times. The calculations were carried out on an NVIDIA GeForce GTX1060 video card. Based on the Python programming language, a system has been developed for measuring accuracy and performance metrics when processing new data.

Practical significance. The developed system, with the expansion of the training set and the corresponding additional training, will significantly reduce the operator's time spent on recognizing pollen grains. The predictions of the proposed model can be used as hints or recommendations for an expert in classifying.

Andriyanov N.A., Kamalova Yu.B. Efficiency and performance study of computer vision algorithms in pollen grain recognition problems. Information-measuring and Control Systems. 2022. V. 20. № 5. P. 48−54. DOI: https://doi.org/10. 18127/j20700814-202205-08 (in Russian)

1. Battiato S., Ortis A., Trenta F., Ascari L., Politi M., Siniscalco C. Detection and classification of pollen grain microscope images. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) Workshops. 2020. № 1. Р. 980-981.
2. Shelekhova T.S., Slukovsky Z.I., Lavrova N.B. Methods for studying bottom sediments of lakes in Karelia [monograph]. Institute of Geology KarRC RAS. Petrozavodsk: KarRC RAS. 2020. 112 p.
3. Chekryga G.P., Nitsievskaia K.N., Yudin O.B. Determination of the botanical origin of honey and the growth of honey-bearing vegetation by the method of pollen analysis. News of higher educational institutions. Food technology. 2021. № 1(379). Р. 94-97.
4. Andrijanov N.A., Dement'ev V.E., Tashlinskij A.G. Obnaruzhenie ob#ektov na izobrazhenii: ot kriteriev Bajesa i Nejmana–Pirsona k detektoram na baze nejronnyh setej EfficientDet. Komp'juternaja optika. 2022. T. 46. № 1. S. 139-159.
   DOI: 10.18287/2412-6179-CO-922 (in Russian).
5. GOST 31769-2012 «Med. Metod opredelenija chastoty vstrechaemosti pyl'cevyh zeren». Jelektronnyj fond pravovoj i normativno-tehnicheskoj dokumentacii [Jelektronnyj resurs]. – Rezhim dostupa: https://docs.cntd.ru/document/1200100244, data obrashhenija 10.09.2022 (in Russian).
6. LeCun Y., Bengio Y. Convolutional networks for images, speech, and time series. The handbook of brain theory and neural networks. 1995. № 3361(10). Р. 1-20.
7. Marshalko D.A., Kubanskih O.V. Arhitektura svertochnyh nejronnyh setej. Uchenye zapiski Brjanskogo gosudarstvennogo universiteta. 2019. № 4(16). S. 10-13 (in Russian).
8. Makarenko A.V. Glubokie nejronnye seti: zarozhdenie, stanovlenie, sovremennoe sostojanie. Problemy upravlenija, 2020. № 2. S. 3-19 (in Russian).
9. Šaulien I., Šukien L., Daunys G., Valiulis G., Vaitkeviˇcius L., Matavulj P., Brdar S., Panic M., Sikoparija B., Clot B., Crouzy B., Sofiev M. Automatic pollen recognition with the Rapid-Eparticle counter: the first-level procedure, experience and next steps. Atmos. Meas. Tech. 2019. № 12. Р. 3435-3452. DOI:10.5194/amt-2018-432.
10. Sevillano V., Aznarte J. L. Improving classification of pollen grain images of the POLEN23E dataset through three different applications of deep learning convolutional neural networks. PloS one. 2018. № 13(9). Р. e0201807. DOI: 10.1371/journal.pone.0201807.
11. Kamalova Ju.B., Andrijanov N.A. Raspoznavanie pyl'cevyh zeren s pomoshh'ju svertochnoj nejronnoj seti VGG-16. Vestnik Juzhno-Ural'skogo gosudarstvennogo universiteta. Ser. Komp'juternye tehnologii, upravlenie, radiojelektronika. 2022. № 22(3). S. 39-46
    (in Russian).
12. Dementyiev V.E., Andriyanov N.A., Vasiliev K.K. Use of Images Augmentation and Implementation of Doubly Stochastic Models for Improving Accuracy of Recognition Algorithms Based on Convolutional Neural Networks. 2020 Systems of Signal Synchronization, Generating and Processing in Telecommunications (SYNCHROINFO). 2020. Р. 1-4. DOI: 10.1109/SYNCHROINFO-49631.2020.9166000.
13. Andrijanov N.A., Kamalova Ju.B., Moiseev G.V. Sravnenie metodov AutoML i ML s uchitelem dlja reshenija zadach raspo-znavanija izobrazhenij. Nejrokomp'jutery: razrabotka, primenenie. 2022. T. 24. № 5. S. 19-27. DOI: https://doi.org/ 10.18127/j19998554-202205-02 (in Russian).
14. Andrijanov N.A., Hasanshin I.Ja., Utkin D.S., Ahmad Aus, Kovylov N.N., Kochkarov A.A. Sistema mashinnogo zrenija dlja opredelenija prostranstvennogo polozhenija ob#ektov na baze YOLOv3 i stereokamery. Nejrokomp'jutery: razrabot-ka, primenenie. 2022. T. 24.
    № 5. S. 74-84. DOI: https://doi.org/ 10.18127/j19998554-202205-08 (in Russian).