Application of artificial intelligence methods to automate post-spawn chum salmon individuals enumeration by means of unmanned aerial vehicles in Khabarovsk Territory

Pacific salmon enumeration by means of unmanned aerial vehicles is promising, but its imple­mentation is associated with the difficulties of manually processing large volumes of aerial imagery in order to count the number of spawners, post-spawn individuals (i.e. spawned out salmon). Automation of counting individuals in unmanned photographic materials by means of artificial intelligence (AI) methods, in particular, using neural networks from deep learning technologies domain, is one of the most promising ways to optimize salmon enumeration by means of unmanned aerial vehicles by increasing its productivity and efficiency. At present, the use of AI to identify individuals in unmanned survey materials has been implemented only for large cartilaginous fish (sharks, rays). This work demonstrates for the first time this possibility for teleosts, namely for post-spawn chum salmon. The proposed approach demonstrates the fundamental possibility of counting relatively small and highly variable in appearance aquatic species, such as Pacific salmon, even in poor visibility in rivers utilized for spawning. The example of the rivers of the Khabarovsk Territory shows that the use of correctly configured (trained) neural networks allows automating the detection and counting of post-spawn chum salmon using unmanned aerial photography materials. A description of the completed creation of a model based on a neural network for solving this problem, implementing the detection and counting of post-spawn salmon using AI methods on a desktop GIS platform is given. The importance of developing the Pacific salmon enumeration using AI methods specifically on a desktop GIS platform is substantiated due to a number of objective advantages of this approach. The proposed automation of post-spawn chum salmon enumeration in the Khabarovsk Territory using neural networks is the beginning of the automation of unmanned enumeration of Pacific salmon using AI methods, taking into account species and regional specifics. The importance of inter-branch interaction of VNIRO divisions for maintaining a common repository of models based on neural networks and their supporting datasets is shown.

1. Бизиков В.А., Петерфельд В.А., Черноок В.И., и др. Методические рекомендации по проведению учёта приплода байкальской нерпы (Pusa sibirica) с беспилотных летательных аппаратов в Байкальском рыбохозяйственном бассейне. М.: Изд-во ВНИРО, 2021. 56 с.

2. Методические рекомендации по учёту численности тихоокеанских лососей в реках Сахалинской области. Южно-Сахалинск: Изд-во СахНИРО, 2013. 31 с.

3. Свиридов В.В., Коцюк Д.В., Подорожнюк Е.В. Беспилотный фотограмметрический учёт тихоокеанских лососей посредством БПЛА потребительского класса // Изв. ТИНРО. 2022а. Т. 202. С. 429–449.

4. Свиридов В.В., Подорожнюк Е.В., Никитин В.Д., Скорик А.В. Модификации беспилотного учёта производителей тихоокеанских лососей в реках Сахалинской области и Хабаровского края // Изв. ТИНРО. 2022б. Т. 202. С. 1015–1031.

5. Attard M.R.G., Phillips R.A., Bowler E., et al. Review of Satellite Remote Sensing and Unoccupied Aircraft Systems for Counting Wildlife on Land // Remote Sensing. 2024. Т. 16. № 4. С. 627.

6. Borowicz A., Le H., Humphries G. et al. Aerial-trained deep learning networks for surveying cetaceans from satellite imagery // PLOS One. 2019. V. 14. № 10. 15 p.

7. Boulent J., Charry B., Kennedy M. et al. Scaling whale monitoring using deep learning: A human-in-the-loop solution for analyzing aerial datasets // Frontiers in Marine Science. 2023. V. 10. 13 p.

8. Butcher P., Colefax A., Gorkin I. et al. The drone revolution of shark science: A review // Drones. 2021. V. 5. № 1. 8 p.

9. Desgarnier L., Mouillot D., Vigliola L. et al. Putting eagle rays on the map by coupling aerial video-surveys and deep learning // Biological Conservation. 2022. V. 267. 24 p.

10. Dujon A., Ierodiaconou D., Geeson J. et al. Machine learning to detect marine animals in UAV imagery: effect of morphology, spacing, behaviour and habitat // Remote Sensing in Ecology and Conservation. 2021. V. 7. № 3. P. 341–354.

11. Gray P., Bierlich K., Mantell S. et al. Drones and convolutional neural networks facilitate automated and accurate cetacean species identification and photogrammetry // Methods in Ecology and Evolution. 2019. V. 10. № 9. P. 1490–1500.

12. Guirado E., Tabik S., Rivas M. et al. Whale counting in satellite and aerial images with deep learning // Scientific reports. 2019. V. 9. № 1. 13 p.

13. Infantes E., Carroll D., Silva W. et al. An automated work-flow for pinniped surveys: a new tool for monitoring population dynamics // Frontiers in Ecology and Evolution. 2022. V. 10. 17 p.

14. Kellenberger B., Veen T., Folmer E., et al. 21 000 birds in 4,5 h: efficient large scale seabird detection with machine learning // Remote Sensing in Ecology and Conservation. 2021. V. 7. № 3. P. 445–460.

15. Rodofili E., Lecours V., LaRue M. Remote sensing techniques for automated marine mammals detection: a review of methods and current challenges // PeerJ. 2022. V. 10. 22 p.