Effective Feature Extraction and Training Sample Generation for Harmful Animal Detection with Limited Image and Audio Data

2022 Fiscal Year Final Research Report

• Project/Area Number: 19K12040
• Research Category: Grant-in-Aid for Scientific Research (C)
• Allocation Type: Multi-year Fund
• Section: 一般
• Review Section: Basic Section 61010:Perceptual information processing-related
• Research Institution: Tokyo University of Agriculture and Technology
• Principal Investigator: Hotta Seiji 東京農工大学, 工学(系)研究科(研究院), 准教授 (90346932)
• Project Period (FY): 2019-04-01 – 2023-03-31
• Keywords: パターン認識 / 機械学習 / 害獣対策 / 動物福祉

Outline of Final Research Achievements

We conducted experimental research to demonstrate that the generation of artificial audio and image patterns repellent to specific harmful animals, can moderately control the behavior of these animals. Additionally, we conducted research on the detection and behavior estimation of free-range chickens from videos, as an application of this study to livestock animals. Similar to harmful animals, the collection of training samples was challenging in this study as well. However, through the application of the artificial training data generation method developed in this research, we experimentally confirmed the potential for improving the accuracy of detection and behavior estimation.

Free Research Field

パターン認識

Academic Significance and Societal Importance of the Research Achievements

本研究は，動物を対象とした機械による害獣駆除や動物福祉のための基盤技術である画像・音声データを利用したパターン認識技術の開発に寄与するものである．具体的には，少数データから害獣をコントロールするための人工音声や視覚パターンを生成する手法と，家畜の検出や行動推定の精度を向上させるための人工パターンを生成する手法を提案し，実験により害獣に対しては行動を制御することが可能であること，および動物福祉に対しては家畜動物の検出や行動認識の精度向上に効果があることを示した．