Artificial Size Slicing Aided Fine Tuning (ASSAFT) and Hyper Inference (ASSAHI) in tomato detection

Abstract

In the realm of precision agriculture, accurate harvest prediction is vital, as any discrepancies between forecasted and actual yields can lead to significant commercial and logistical challenges. This paper presents a novel deep learning-based approach for detecting and counting tomato fruits using advanced computer vision techniques. Building upon our previously established framework for ultra-wide image acquisition, this approach focuses on a unique patch-cropping technique tailored to tomatoes. This method aligns with the natural clustering of tomatoes, significantly improving object detection in greenhouse settings and thereby enhancing the model’s performance in identifying individual fruits. The detection results exhibit a precision of 0.85, a recall of 0.93, and an F1-score of 0.89. Our approach’s efficacy is also demonstrated through a case study on harvest prediction in a tomato greenhouse. The proposed methodology exhibited a lower error rate than the agronomist’s estimates and proved its practical applicability. These findings suggest that our methodology could substantially contribute to optimizing sustainable farming practices, offering a promising direction for future research and application in the agricultural sector.

Plant phenotyping relevance

トマト果実の検出・計数という植物器官形質の取得を目的に、画像取得とパッチクロッピングを含む深層学習手法を開発・評価しており、方法が中心的である。

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.