Non-Destructive Watermelon Ripeness Assessment Using Deep Learning and Field-Based RGB Imagin

Abstract

Accurate, non-destructive assessment of watermelon ripeness remains a significant challenge in horticultural production, particularly under field conditions where traditional visual and tactile evaluation methods are subjective and often inconsistent. Although mechanical, acoustic, and spectroscopic techniques have demonstrated promising performance, their reliance on controlled laboratory environments limits their practical applicability in real-world agricultural settings. This study presents a field-deployable, AI-assisted computer vision system designed for objective, real-time classification of watermelon ripeness. The proposed prototype combines controlled illumination with RGB imaging and convolutional neural networks trained on thousands of annotated outdoor images collected over multiple growing seasons. A phased development strategy—encompassing proof-of-concept modelling, field integration, and multi-season validation—supports robustness against variable lighting conditions and environmental influences. The anticipated outcome is a reliable, non-destructive decision-support tool for growers, capable of identifying ripe fruit for manual harvesting while providing a technological foundation for future autonomous harvesting and precision agriculture applications.

Plant phenotyping relevance

スイカ果実の成熟度という植物器官の状態を、RGB画像とCNNで非破壊推定する手法を開発し、圃場統合と複数季節の検証まで行うため、フェノタイピング手法が中心です。

Code and data availability

公開論文であることは確認できましたが、現在の公式API・許可済み取得経路では本文を自動取得できませんでした。

No evidence-backed public reproduction asset is currently recorded.