Lightweight attention enhanced YOLOv11 for accurate multi class detection of brinjal diseases.

Abstract

The cultivation of brinjal is seriously affected by various pathogens which affect crop yield and quality hence the importance of automated disease detection systems in precision agriculture. The conventional image-based diagnostic methods fail to differentiate between similar diseases classes that look similar and this results in misdiagnosis and poor crop management. This research paper advances to suggest a lightweight deep learning model, IEMA-YOLOv11 (Improved Efficient Multi-Scale Attention-based YOLOv11), to successfully identify and detect Brinjal disease successfully in real-time. The model combines an EMA (Efficient Multi-Scale Attention) system, IRMB (Inverted Residual Mobile Blocks), and a LDFM (Local Detail Feature Module) to obtain fine-grained lesion features, as well as a refined MPDIoU loss to achieve a better object localization. The framework was tested against a curated multi-class brinjal disease data of six fungal, two bacterial, three viral and one nematode infection. IEMA-YOLOv11 had a precision of 96.5, a recall of 95.7, mAP@50 of 95.6 and mAP@50:95 of 94.9, with only 5.12M parameters and 17.1 GFLOPs. In comparison to the current benchmarks, IEMA-YOLOv11 was superior to YOLOv8-n by a margin of 2.4% in mAP@50, which has a potential to be used on a large scale to detect brinjal disease in a sustainable manner.

Plant phenotyping relevance

植物病害の病斑特徴を画像から抽出・検出する深層学習手法を開発し、データセット上で性能検証しているため、植物フェノタイピング手法が研究の中心である。

Code and data availability

The paper's brinjal disease dataset (5,850 images, 12,767 YOLO annotations) and IEMA-YOLOv11 analysis assets are not publicly deposited; the Data Availability Statement says they are available from the corresponding author on reasonable request. No public repository, code URL, or trained model checkpoint is provided in

No evidence-backed public reproduction asset is currently recorded.