Lightweight Vision–Transformer Network for Early Insect Pest Identification in Greenhouse Agricultural Environments

TL;DR

This paper introduces a lightweight AI model for quickly and accurately identifying pests and diseases in greenhouse crops, even under tough conditions.

Contribution

Light-HortiNet is a novel lightweight model combining a Mobile-Transformer backbone with attention and small-object enhancement for efficient greenhouse pest detection.

Findings

Light-HortiNet outperforms existing lightweight models in detection accuracy and small-target recognition in greenhouse environments.

The model achieves real-time inference speeds over 20 FPS on edge devices like Jetson Nano.

It shows strong performance with mAP@50 of 0.872 and high classification accuracy of 0.894 on a greenhouse pest and disease dataset.

Abstract

This paper presents Light-HortiNet, a lightweight deep learning model tailored for real-time detection of pests and diseases in greenhouse-grown fruits and vegetables. Designed for edge devices like the Jetson Nano, it combines a Mobile-Transformer backbone with cross-scale attention and small-object enhancement techniques to achieve high accuracy under challenging greenhouse conditions—such as variable lighting and high humidity—while maintaining fast inference speeds (over 20 FPS). Experiments show it outperforms existing lightweight models in both overall detection performance and small-target recognition, making it well-suited for practical deployment in smart horticulture systems.

This study addresses the challenges of early recognition of fruit and vegetable diseases and pests in facility horticultural greenhouses and the difficulty of real-time deployment on edge devices, and proposes a lightweight cross-scale intelligent recognition network, Light-HortiNet, designed to achieve a balance between high accuracy and high efficiency for automated greenhouse pest and disease detection. The method is built upon a lightweight Mobile-Transformer backbone and integrates a cross-scale lightweight attention mechanism, a small-object enhancement branch, and an alternative block distillation strategy, thereby effectively improving robustness and stability under complex illumination, high-humidity environments, and small-scale target scenarios. Systematic experimental evaluations were conducted on a greenhouse pest and disease dataset covering crops such as tomato, cucumber, strawberry, and pepper. The results demonstrate significant advantages in detection performance, with mAP@50 reaching 0.872, mAP@50:95 reaching 0.561, classification accuracy reaching 0.894, precision reaching 0.886, recall reaching 0.879, and F1-score reaching 0.882, substantially outperforming mainstream lightweight models such as YOLOv8n, YOLOv11n, MobileNetV3, and Tiny-DETR. In terms of small-object recognition capability, the model achieved an mAP-small of 0.536 and a recall-small of 0.589, markedly enhancing detection stability for micro pests such as whiteflies and thrips as well as early-stage disease lesions. In addition, real-time inference performance exceeding 20 FPS was achieved on edge platforms such as Jetson Nano, demonstrating favorable deployment adaptability.

Figures