Log physical appearance quantitatively analyzed by image analysis and its relationship with log strength properties: toward log strength estimation using machine learning model.

2021 Fiscal Year Final Research Report

• Project/Area Number: 19K06321
• Research Category: Grant-in-Aid for Scientific Research (C)
• Allocation Type: Multi-year Fund
• Section: 一般
• Review Section: Basic Section 41040:Agricultural environmental engineering and agricultural information engineering-related
• Research Institution: Kyoto Prefectural University
• Principal Investigator: Nagashima Keiko 京都府立大学, 生命環境科学研究科, 教授 (40582987)
• Co-Investigator(Kenkyū-buntansha): 杜 偉薇 京都工芸繊維大学, 情報工学・人間科学系, 准教授 (00512790) ; 神代 圭輔 京都府立大学, 生命環境科学研究科, 准教授 (00548448)
• Project Period (FY): 2019-04-01 – 2022-03-31
• Keywords: 木口の断面画像 / 画像解析 / 年輪情報 / 動的ヤング率 / 原木の強度推定 / 強度等級 / 機械学習

Outline of Final Research Achievements

By applying image processing and machine learning, this study aims to establish a method to detect wood features（ex. number of annual rings and average width of annual rings), to interpret the relationship between the wood features and wood strength, and to estimate the wood strength based on their relationships. To accurately measure the wood annual ring information by image processing, we proposed a new model combining the total variation algorithm, Hough transform and Convolutional Neural Network and showed its efficiency on indoor wood images. However, challenges remain for outdoor images. Investigating the relationship between annual ring information and wood strength, the woods became stronger as the number of annual rings became larger and the average width of annual rings,15th rings from the center and outside became smaller. Based on this relationship, a SVM model was established and showed its efficiency on detecting strong woods with a relatively high precision (62.5%).

Free Research Field

森林計画学

Academic Significance and Societal Importance of the Research Achievements

本研究で構築した強度推定モデルは年輪情報から強度等級が高い原木を6割の確率で推定できた。これは、強度が高い原木を 100 本購入したい場合、無作為では433 本購入するところ、予測モデルを使用すれば 160 本 の購入で済むこととなり、現在課題となっている適切な素材供給と効率的な素材確保が可能になることを示した。また、本研究のように年輪情報からの強度推定を試みた研究はこれまでになく、その有効性を示した本研究の学術的意義は大きい。画像解析による年輪情報の抽出には課題は残るが、今後の技術改良により画像を用いた強度推定による原木仕分けの自動化、生産性・収益性の向上に貢献すると期待される。