Intelligent deep learning architecture for precision vegetable disease detection advancing agricultural new quality productive forces.

Abstract

In the context of advancing agricultural new quality productive forces, addressing the challenges of uneven illumination, target occlusion, and mixed infections in greenhouse vegetable disease detection becomes crucial for modern precision agriculture. To tackle these challenges, this study proposes YOLO-vegetable, a high-precision detection algorithm based on improved You Only Look Once version 10 (YOLOv10). The framework incorporates three innovative modules. The Adaptive Detail Enhancement Convolution (ADEConv) module employs dynamic parameter adjustment to preserve fine-grained features while maintaining computational efficiency. The Multi-granularity Feature Fusion Detection Layer (MFLayer) improves small target localization accuracy through cross-level feature interaction mechanisms. The Inter-layer Dynamic Fusion Pyramid Network (IDFNet) combines with Attention-guided Adaptive Feature Selection (AAFS) mechanism to enhance key information extraction capability. Experimental validation on our self-built Vegetable Disease Dataset (VDD, 15,000 images) demonstrates that YOLO-vegetable achieves 95.6% mean Average Precision at IoU threshold 0.5, representing a 6.4 percentage point improvement over the baseline model. The method maintains efficiency with 3.8M parameters and 18.6ms inference time per frame, providing a practical solution for intelligent disease detection in facility agriculture and contributing to the development of agricultural new quality productive forces.

Plant phenotyping relevance

植物の病徴・病害状態を画像から検出する深層学習手法を開発し、自作データセットで性能検証しており、植物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.