Lightweight attention-enhanced EfficientDet for tomato leaf detection and disease classification in real-world conditions
- 1 September 2026
- journal article
- Published by IOP Publishing in Engineering Research Express
- Vol. 8 (18) , 185213
Abstract
Tomatoes are a critical crop for global food security, offering substantial nutritional and economic value. Their widespread availability makes them a key component in combating malnutrition. However, diseases affecting tomato plants significantly threaten agricultural productivity. While numerous deep learning-based approaches have been proposed to address this challenge, many rely on datasets collected in controlled environments. As a result, these models often underperform in real-field conditions due to complex backgrounds and variable lighting. Accurate leaf detection is a crucial first step in enabling automated disease monitoring under such conditions, as it facilitates precise downstream analysis. In this study, we present Lite-AttnEffDet, a lightweight and efficient leaf detection model built upon EfficientDet and enhanced with lightweight attention mechanisms. Specifically, we incorporate an efficient channel attention module into the detection head to strengthen channel-wise feature emphasis, thereby improving leaf localization in cluttered, real-world scenes. The proposed Lite-AttnEffDet was evaluated on the FieldPlant dataset, which contains tomato plant images captured in real agricultural settings. Experimental results demonstrate that Lite-AttnEffDet achieves a mAP@IoU = 0.50 of 0.930 ± 0.014 and a mean Recall@IoU = 0.50 of 0.860 ± 0.021 over five independent runs, outperforming the baseline EfficientDet-D0 and other conventional detectors while requiring only 2.6 GFLOPs and increasing the number of trainable parameters by just 2.56% over the baseline. The improved leaf detection capability directly benefits downstream classification, where convolutional neural network-based models achieve up to 88% accuracy, a notable improvement over the 82% accuracy observed when using raw images. These results underscore the robustness of Lite-AttnEffDet in real-world field environments, making it a promising solution for plant health monitoring and precision agriculture.Keywords
This publication has 29 references indexed in Scilit:
- SqueezeNet-Based Deep Learning Framework for Accurate Tomato (Solanum lycopersicum) Leaf Disease Diagnosis and ClassificationInternational Journal of Computational Intelligence Systems, 2025
- CONF-RCNN: a conformer and faster region-based convolutional neural network model for multi-label classification of tomato leaves disease in real field environmentJournal of Plant Diseases and Protection, 2025
- TomSevNet: a hybrid CNN model for accurate tomato disease identification with severity level assessmentNeural Computing and Applications, 2023
- Improving Deep Learning-based Plant Disease Classification with Attention MechanismGesunde Pflanzen, 2022
- A comprehensive study of feature extraction techniques for plant leaf disease detectionMultimedia Tools and Applications, 2021
- Attention embedded residual CNN for disease detection in tomato leavesApplied Soft Computing, 2020
- A comparative study of fine-tuning deep learning models for plant disease identificationComputers and Electronics in Agriculture, 2019
- Impact of dataset size and variety on the effectiveness of deep learning and transfer learning for plant disease classificationComputers and Electronics in Agriculture, 2018
- Deep learning models for plant disease detection and diagnosisComputers and Electronics in Agriculture, 2018
- Major Tomato Viruses in the Mediterranean Basin2012