Evaluation of the ability to measure morphological structures of plants obtained from tissue culture applying image processing techniques

Abstract

Abstract Biotechnological approaches, for instance, plant tissue culture, can be used to improve and accelerate the reproduction of plants. A single portion of a plant can produce many plants throughout the year in a relatively short period of laboratory conditions. Monitoring and recording plant morphological characteristics such as root length and shoot length in different conditions and stages are necessary for tissue culture. These features were measured using graph paper in a laboratory environment and sterile conditions. This research investigated the ability to use image processing techniques in determining the morphological features of plants obtained from tissue culture. In this context RGB images were prepared from the plants inside the glass, and different pixel-based and object-based classification methods were applied to an image as a control. The accuracy of these methods was evaluated using the kappa coefficient, and overall accuracy was obtained from Boolean logic. The results showed that among pixel-based classification methods, the maximum likelihood method with a kappa coefficient of 87% and overall accuracy of 89.4 was the most accurate, and the Spectral angle mapper method (SAM) method with a kappa coefficient of 58% and overall accuracy of 54.6 was the least accurate. Also, among object-based classification methods, Support Vector Machine (SVM), Naïve Bayes, and K-nearest neighbors algorithm (KNN) techniques, with a Kappa coefficient of 88% and overall accuracy of 90, can effectively distinguish the cultivation environment, plant, and root. Comparing the values of root length and shoot length estimated in the laboratory culture environment with the values obtained from image processing showed that the use of the SVM image classification method, which is capable of estimating root length and shoot length with RMSE 2.4, MAD 3.01 and R2 0.97, matches the results of manual measurements with even higher accuracy.

Plant phenotyping relevance

組織培養植物の根長・シュート長を画像処理で推定する手法を開発・比較し、手動測定との精度検証まで行っており、表現型取得が研究の中心である。

Code and data availability

The paper's plant images, classification results, and measurement datasets are not publicly deposited; the Data Availability Statement states they are available only from the corresponding author on reasonable request. No author analysis code, models, or public repository URL is provided.

No evidence-backed public reproduction asset is currently recorded.