Lightweight maturity detection of greenhouse tomato based on YO-LO-LTD

LI Quanwu; YANG Beibei; MEI Fengtong; TANG Yuan

doi:10.16861/j.cnki.zggc.2025.0027

Abstract：To address the issues of missed and false detections caused by complex backgrounds and scale variations in tomato fruit maturity detection,as well as the limitations of existing methods in terms of efficiency and deployment,this study proposes a lightweight greenhouse tomato maturity detection algorithm based on YOLO-LTD.Building upon YO-LOv11-n as the baseline,the model introduces the following innovations:(1)A cross-attention module is incorporated into the backbone network to mitigate the interference of occlusions between leaves,stems,and fruits on detection accura-cy,thereby enhancing feature extraction capabilities for key regions.(2)The lightweight GSConv module replaces stan-dard convolutions in the neck network,optimizing computational efficiency while preserving feature representation,and reducing both model parameter count and computational complexity.(3)An adaptive spatial feature fusion module is embedded in the head network to alleviate inconsistencies between multi-scale features,further improving robustness and generalization.Experimental results demonstrate that YOLO-LTD achieves a mean average precision(mAP),recall,and accuracy of 94.23％,95.44％,and 92.07％,respectively,with an inference time of 7.21 ms and a compact model size of 5.18 Mb.Compared to YOLOv11-n,YOLO-LTD improves mAP,recall,and accuracy by 2.50 percentage points,2.80 per-centage points,and 1.60 percentage points,respectively,while exhibiting higher efficiency and smaller model size.When evaluated against Mask R-CNN,Faster R-CNN,and other YOLO variants,YOLO-LTD demonstrates superior perfor-mance in both accuracy and efficiency,highlighting its potential for widespread application in greenhouse environments.This research provides a theoretical foundation and technical support for orchard yield estimation,crop growth monitor-ing,cultivation optimization,and the development of tomato-picking robots.