Detection of color phenotype in strawberry germplasm resources based on field robot and semantic segmentation

Abstract

Strawberry holds significant economic value, but the laborious and time-consuming process of evaluating phenotypic traits in numerous germplasm resources during breeding poses a challenge. Prior studies relied on manual image collection within a single laboratory background, making it difficult to achieve automatic image collection and precise segmentation in complex field environments. However, accurate segmentation of plant organs is crucial for reliable phenotyping. In this research, we collected strawberry images at three growth stages (vegetative, flowering, and fruiting) using mobile phones and a high-resolution industrial camera mounted on our self-developed robot. Next, we designed an improved semantic segmentation model specifically tailored for strawberry plants, named Strawberry Segment Model (SSM), based on the Segment Anything Model. To address the uneven sample distribution problem, we enhanced the loss function and introduced a multi-loss approach combined with the class weight, resulting in improved detection performance. The comparative results demonstrated that SSM achieved state-of-the-art segmentation performance on the mobile phone image set, with a mean Intersection over Union (mIoU) of 80.20 %. We updated the model for the industrial camera on the robot, and achieved 75.81 % mIoU with only 10 % of the new data, striking a balance between performance and cost. Additionally, we mitigated uneven illumination using the Contrast Limited Adaptive Histogram Equalization and employed a Support Vector Machine model to classify 90 germplasm resources. The accuracy rates were 100 % for leaves and flowers, and 92.59 % for fruits. Overall, this study introduces novel equipment and methods for automated phenotypic analysis, supporting breeding investigations.

Plant phenotyping relevance

イチゴの生殖質を対象に、ロボット搭載カメラによる画像取得、器官セグメンテーション、色表現型の分類を開発・評価しており、表現型取得手法が研究の中心である。

Code and data availability

公開状態または取得可能な本文経路を確認できませんでした。

No evidence-backed public reproduction asset is currently recorded.