From architectural complexity to minimalist optimization: refining microscopic tea disease detection via Opti-YOLOv11n.

Abstract

Introduction The precise detection of microscopic tea leaf diseases is a prerequisite for sustainable precision agriculture. While recent deep learning advancements often favor architectural complexity, this "complexity bias" frequently introduces computational redundancy-a "complexity tax"-that destabilizes gradient flow and fails to resolve critical resolution bottlenecks for micro-lesion identification. Methods We propose Opti-YOLOv11n, a minimalist optimization paradigm prioritizing physical input fidelity and stabilized gradient dynamics. Utilizing a dataset of six pathological categories, our framework employs high-resolution scaling (832×832) combined with a momentum-based SGD optimizer and a cosine annealing schedule to reconstruct essential spatial textures. Results Opti-YOLOv11n achieved a peak Precision of 98.87% and a Recall of 95.97%, while reducing the parameter count to 2.35 M-a 9.2% decrease relative to the baseline-and maintaining a real-time inference speed of 104.7 FPS on edge-simulated hardware. Discussion Statistical verification via 5-fold cross-validation confirms superior generalization stability. These results substantiate that strategic structural pruning and physical input scaling provide a more robust technical benchmark for autonomous plant protection than the adoption of excessive architectural depth.

Plant phenotyping relevance

茶葉病害の微小病斑を画像から検出する深層学習手法を開発し、精度・計算量・推論速度・交差検証で技術評価しているため、植物病害フェノタイピング手法が中心である。

Code and data availability

植物フェノタイピング解析を再現する公開資産であることを、入力本文と直接リンクから確認できなかったため保留しました。

No evidence-backed public reproduction asset is currently recorded.