講演抄録/キーワード |
講演名 |
2023-03-03 09:30
[ショートペーパー]スピリット学習を用いたUAV空撮画像分散学習システムの性能評価 ○ソン テイカイ・王 瀟岩(茨城大)・梅比良正弘(南山大) SR2022-93 |
抄録 |
(和) |
インフラ点検や被災者支援などを効率的に行うため、UAV(無人航空機)の空撮画像から自動的にオブジェクトを認識する手法が近年注目されている。従来の集中学習や連合学習を用いた手法は、それぞれ画像のプライバシーとUAV計算負荷の懸念がある。本研究では、スプリット学習を用いたUAV空撮画像分散学習システムを提案し、シミュレーションを行った。その結果、学習時間が短縮でき、特にnon-IIDデータの場合に効果があることがわかった。 |
(英) |
Due to its ease of deployment and high mobility, unmanned aerial vehicles (UAVs) have gained popularity for a variety of applications. Coordinating the actions of several UAVs, however, can be difficult when completing high-level, complicated tasks like search and rescue missions, target surveillance, and information dissemination.
To address this issue, distributed learning methods such as federated learning (FL) and split learning (SL) have been proposed. FL involves building a joint model by aggregating models trained on each UAV's local data, while SL involves splitting the model between the UAVs and a central server, with both parties collaborating to train the entire network. In this paper, investigations were made on the application of split learning (SL) in a multi-UAV system for image classification in scenarios including area exploration. A server was used to coordinate multiple UAVs, with each UAV using a local model trained on images gathered by its on-board camera to perform classification tasks. Local updates from all UAVs were communicated to the server, which then performed a global update and transmitted the results back to the UAVs. The performance of the proposed system was evaluated using an aerial perspective geographic dataset, and the effectiveness of SL compared to federated learning (FL) was discussed. It was found that SL significantly reduces local computation compared to FL, leading to faster learning times, and is particularly effective with unbalanced data. It also requires less data during the initial phases of training and has a faster convergence speed compared to centralized learning. |
キーワード |
(和) |
分散学習 / 無人航空機 / 畳み込みニューラルネットワーク / 分裂学習 / 連合学習 / / / |
(英) |
distributed machine learning / Unmanned aerial vehicles / convolutional neural network / split learning / federated learning / / / |
文献情報 |
信学技報, vol. 122, no. 400, SR2022-93, pp. 44-46, 2023年3月. |
資料番号 |
SR2022-93 |
発行日 |
2023-02-22 (SR) |
ISSN |
Online edition: ISSN 2432-6380 |
著作権に ついて |
技術研究報告に掲載された論文の著作権は電子情報通信学会に帰属します.(許諾番号:10GA0019/12GB0052/13GB0056/17GB0034/18GB0034) |
PDFダウンロード |
SR2022-93 |
|