U-FedTomAtt: Ultra-lightweight Federated Learning with Attention for Tomato Disease Recognition

Romiyal George; Sathiyamohan Nishankar; Selvarajah Thuseethan; Chathrie Wimalasooriya; Yakub Sebastian; Roshan G. Ragel; Zhongwei Liang

q-bio.QMFeb 18, 2026

U-FedTomAtt: Ultra-lightweight Federated Learning with Attention for Tomato Disease Recognition

Authors: Romiyal George, Sathiyamohan Nishankar, Selvarajah Thuseethan, Chathrie Wimalasooriya, Yakub Sebastian, Roshan G. Ragel, Zhongwei Liang

Organizations: Department of Computer Engineering, University of Peradeniya, Sri Lanka. · Faculty of Science and Technology, Charles Darwin University, Australia. · Canadian Institute for Cybersecurity, University of New Brunswick, Canada. · School of Mechanical and Electrical Engineering, Guangzhou University, China.

Abstract

Federated learning has emerged as a privacy-preserving and efficient approach for deploying intelligent agricultural solutions. Accurate edge-based diagnosis across geographically dispersed farms is crucial for recognising tomato diseases in sustainable farming. Traditional centralised training aggregates raw data on a central server, leading to communication overhead, privacy risks and latency. Meanwhile, edge devices require lightweight networks to operate effectively within limited resources. In this paper, we propose U-FedTomAtt, an ultra-lightweight federated learning framework with attention for tomato disease recognition in resource-constrained and distributed environments. The model comprises only 245.34K parameters and 71.41 MFLOPS. First, we propose an ultra-lightweight neural network with dilated bottleneck (DBNeck) modules and a linear transformer to minimise computational and memory overhead. To mitigate potential accuracy loss, a novel local-global residual attention (LoGRA) module is incorporated. Second, we propose the federated dual adaptive weight aggregation (FedDAWA) algorithm that enhances global model accuracy. Third, our framework is validated using three benchmark datasets for tomato diseases under simulated federated settings. Experimental results show that the proposed method achieves 0.9910% and 0.9915% Top-1 accuracy and 0.9923% and 0.9897% F1-scores on SLIF-Tomato and PlantVillage tomato datasets, respectively.

Figures & tables

Fig. 1: Illustration of the proposed ultra-lightweight neural network with LoGRA for tomato leaf disease recognition. The red arrows denote residual connections, the blue arrows represent standard connections, and the green dashed lines indicate the expansion of the layer architecture.

Fig. 2: Illustration of FedDAWA algorithm.

#	Layer	Input Shape	[Kernel Size, Stride]	Padding/Dilation	Activation	Output Shape
1	Conv2D	256 $\times$ 256	[3 $\times$ 3, 2]	padding=1	ReLU	128 $\times$ 128
2	DBNeck /wo R	128 $\times$ 128	[1 $\times$ 1 $\rightarrow$ 3 $\times$ 3 $\rightarrow$ 1 $\times$ 1, 2]	dilation=1	ReLU	64 $\times$ 64
3	DBNeck /wo R	64 $\times$ 64	[1 $\times$ 1 $\rightarrow$ 3 $\times$ 3 $\rightarrow$ 1 $\times$ 1, 2]	dilation=1	ReLU	32 $\times$ 32
4	DBNeck /w R	32 $\times$ 32	[1 $\times$ 1 $\rightarrow$ 3 $\times$ 3 $\rightarrow$ 1 $\times$ 1, 1]	dilation=1	ReLU	32 $\times$ 32
5	DBNeck /wo R	32 $\times$ 32	[1 $\times$ 1 $\rightarrow$ 3 $\times$ 3 $\rightarrow$ 1 $\times$ 1, 2]	dilation=2	ReLU	16 $\times$ 16
6	DBNeck /w R	16 $\times$ 16	[1 $\times$ 1 $\rightarrow$ 3 $\times$ 3 $\rightarrow$ 1 $\times$ 1, 1]	dilation=1	ReLU	16 $\times$ 16

TABLE I: Detailed configuration of the proposed ultra-lightweight neural network with LoGRA.

Fig. 3: The selection of the hyperparameter

\beta

for SLIF-Tomato and PlantVillage tomato datasets.

	Model	Local 1		Local 2		Local 3		Local 4		Local 5		FL (Global)
	Model	Top-1	F1	Top-1	F1	Top-1	F1	Top-1	F1	Top-1	F1	Top-1	F1
SLIF-Tomato	MobileNetV2 [ 39 ]	.9961	.9966	.9843	.9867	.9412	.9467	.9485	.9563	.7890	.8046	.9731	.9775
	SqueezeNet1_0 [ 40 ]	.9709	.9745	.9569	.9615	.9692	.9728	.9351	.9434	.9754	.9784	.9748	.9776
	ShuffleNetV2 [ 41 ]	.9759	.9797	.9894	.9910	.9737	.9772	.9743	.9782	.9933	.9943	.9771	.9821
	PSLT_tiny [ 38 ]	.9457	.9493	.9855	.9866	.9715	.9732	.9838	.9857	.8987	.9035	.9838	.9863
	PiT-tiny [ 37 ]	.9933	.9940	.9843	.9867	.9944	.9949	.9172	.9285	.9737	.9772	.9799	.9816

TABLE II: Federated performance comparison (Top-1 Accuracy and F1-score) between the proposed approach and existing state-of-the-art deep networks. The proposed FedDAWA FL aggregation is applied for all networks compared.

Fig. 4: Evolution of local and global model accuracies over 50 communication rounds. The continuous lines and broken lines indicate the accuracy of the global model and local models, respectively.

Model	Efficiency Metrics
Model	FLOPs (M)	Params (K)
MobileNetV2 [ 39 ]	399.95	2,234.12
SqueezeNet1_0 [ 40 ]	973.30	739.53
ShuffleNetV2 [ 41 ]	390.14	2,486.82
PSLT_tiny [ 38 ]	874.40	3,917.07
PiT-tiny [ 37 ]	652.70	4,607.18

TABLE III: Comparison of computational efficiency of the proposed ultra-lightweight neural network against existing state-of-the-art neural networks.

Method	SLIF-Tomato				PlantVillage Tomato
Method	Top-1	F1	Precision	Recall	Top-1	F1	Precision	Recall
Non-FL	.9407	.9242	.9242	.9201	.9127	.9208	.9205	.9255
+ FedAvg [ 12 ]	.9873	.9843	.9851	.9837	.9821	.9853	.9845	.9869
+ FedProx [ 13 ]	.9879	.9860	.9881	.9841	.9849	.9871	.9858	.9889
+ FedNova [ 14 ]	.9865	.9834	.9865	.9807	.9855	.9881	.9871	.9896
+ FeDABoost [ 42 ]	.9851	.9808	.9826	.9792	.9815	.9846	.9837	.9862

TABLE IV: Performance comparison of federated aggregation methods and the baseline non-FL approach against the proposed FedDAWA , using the proposed ultra-lightweight neural network.

Lightweight Neural Networks

See all 18 papers

CoAtNet-DeepMoE: A Convolution-Attention Hybrid with DeepSeek Mixture-of-Experts for Parameter-Efficient Tomato Disease Classification
Sep 16, 2026Md Nadim Mahamood, Md Arif Shahriar, Md Shafi Ud Doula +1Plant Disease Classification Hybrid CNN-Transformer Architectures
Department of Computer Science and Engineering Begum Rokeya University Rangpur, Bangladesh · Department of Computer Science Texas State University San Marcos, Texas, USA · Department of Information and Communication Technologies Asian Institute of Technology Pathum Thani, Thailand
AgriKD: Cross-Architecture Knowledge Distillation for Efficient Leaf Disease Classification
May 2, 2026Minh-Dung Le, Minh-Duc Hoang, Hoang-Vu Truong +1Efficient Neural Network Inference Plant Disease Classification
AIT laboratory, Faculty of Artificial Intelligence, FPT University, Da Nang, 550000, Viet Nam
FecalFed: Privacy-Preserving Poultry Disease Detection via Federated Learning
Apr 1, 2026Tien-Yu ChiClassification Federated Learning
Independent Researcher

Sep 16, 2026J/K move · Enter open · S save

CoAtNet-DeepMoE: A Convolution-Attention Hybrid with DeepSeek Mixture-of-Experts for Parameter-Efficient Tomato Disease Classification

Md Nadim Mahamood, Md Arif Shahriar, Md Shafi Ud Doula, Kamrul Hasan

Department of Computer Science and Engineering Begum Rokeya University Rangpur, Bangladesh · Department of Computer Science Texas State University San Marcos, Texas, USA · Department of Information and Communication Technologies Asian Institute of Technology Pathum Thani, Thailand

Plant Disease Classification Hybrid CNN-Transformer Architectures

Open paper AskPDF arXiv

The world population is growing rapidly, and technology is improving in parallel. Meeting the huge demand for food for these 7 billion people not only depends on increasing food production but also on reducing food loss. Crop losses due to disease affect both the food supply and the financial and economic stability of a country. Tomatoes are among the top food-producing crops globally, and a significant portion of this production is lost due to disease. People have used Machine Learning techniques for feature extraction and early diagnosis of tomato diseases, and nowadays, Deep Learning-based models are widely used for disease recognition. However, most existing models are highly parameter-intensive, which increases the time required for training and inference. As a result, while lightweight models are more suitable for user-friendly applications, they often show a reduction in performance. To balance performance and model size, we propose CoAtNet-DeepMoE, a Convolution-Attention hybrid architecture for rich feature extraction, further enhanced with a DeepSeek Mixture of Experts to substantially reduce the number of parameters without sacrificing accuracy. We evaluate our model on both balanced and imbalanced datasets from Kaggle and PlantVillage, demonstrating robustness and achieving 99.80% accuracy, 99.80% precision, 99.80% recall, and 99.80% F1-score on Kaggle, and 99.83% accuracy, 99.85% precision, 99.76% recall, and 99.80% F1-score on PlantVillage, representing state-of-the-art performance with only 2.47M parameters. The source code will be available at https://github.com/nadimbrur/CoAt-MoE.

Abstract

Lightweight Neural Networks

Explore similar work

CoAtNet-DeepMoE: A Convolution-Attention Hybrid with DeepSeek Mixture-of-Experts for Parameter-Efficient Tomato Disease Classification

AgriKD: Cross-Architecture Knowledge Distillation for Efficient Leaf Disease Classification

FecalFed: Privacy-Preserving Poultry Disease Detection via Federated Learning