AUTOMATED DETECTION OF MACROBENTHOS IN TIDAL FLATS USING UNMANNED AERIAL VEHICLES AND DEEP LEARNING

Monitoring Brachyura in tidal flats is labor intensive and limited by the difficulty of accessing such environments. An unmanned aerial vehicle and a deep-learning algorithm were used for image collection in our method for automatically detecting and classifying Brachyura in tidal flats. Our target area was the Baramarae tidal flats in Taean-gun, Republic of Korea, which are part of the Taeanhaean National Park. The target species were mainly the endangered species Austruca lactea, Tubuca arcuata, Macrophthalmus japonicus, and Scopimera globosa. We selected the U-Net convolutional neural network to classify Brachyura species. After training the U-net model and conducting an accuracy evaluation, the automatic identification and classification of different Brachyura species were performed. The overall precision and recall rates were >80. The verification process enabled an effective estimation of the population of each Brachyura species.