In response to challenges in land cover classification (LCC), many researchers have experimented recently with classification methods based on artificial intelligence techniques. For LCC mapping of the vegetated Asahi River in Japan, the current study uses deep learning (DL)-based DeepLabV3+ module for image segmentation of aerial photographs. We modified the existing model by concatenating data on its resultant output port to access the airborne laser bathymetry (ALB) dataset, including voxel-based laser points and vegetation height (i.e. digital surface model data minus digital terrain model data). Findings revealed that the modified approach improved the accuracy of LCC greatly compared to our earlier unsupervised ALB-based method, with 25 and 35% improvement, respectively, in overall accuracy and the macro F1-score for November 2017 dataset (no–leaf condition). Finally, by estimating flow-resistance parameters in flood modelling using LCC mapping-derived data, we conclude that the upgraded DL methodology produces better fit between numerically analyzed and observed peak water levels.

河川を遡上する稚アユの個体数の把握はアユの資源量調査において重要である．しかし，現在は水中カメラ画像を長時間かけて目視確認することにより計数を行っている．本研究では岡山県吉井川の鴨越堰の魚道出口に設置した水中カメラの画像を利用してデータセットを作成し，物体検出の深層学習アルゴリズムであるYOLOv5を用いることで，遡上する稚アユの検出モデルを作成した．モデルによる検出の結果，アユのデータ数の増加にともない精度が向上したが，アユと誤判定されやすい対象物へのラベリングでは，データ数が少数の場合は精度が低下した．本研究のモデルにより，目視計数をすべき画像枚数は従来の1 %弱に抑えられ，また，モデルで画像を判別する場合，約2か月分の画像処理で，従来の手法より所要時間が約90時間短縮されると試算された．

In recent years, the dumping of garbage in rivers has become a common occurrence and has gradually started to affect the normal flow of river channels, which has added lots of work to the river patrol staff. Facing these problems, river authorities urgently need a reasonable and better cost-performance method, that can be adopted on a large scale to support the staff in investigating the garbage within the rivers. Although object detection using Artificial Intelligence has its advantages, it has not been widely applied in the riverine environment using drone. This study attempts to detect garbage in the Asahi River, Japan using two object detection models. By using a large amount of PET images collected from the Internet as training dataset and experimenting with a variety of model-related parameters (i.e., Batch size, Epochs), this study achieved high-accuracy results in recognition of the garbage in the study site. Conclusively, the additional dataset of PET for training, with the similar GSD as test dataset, can improve the Recall value. Nevertheless, without combining with Original dataset collected from the study site, it is difficult to detect the PET using additional dataset only. Thus, combination of Original and additional dataset is a relatively better method to improve the Recall value of detecting PET.