Deep Learning Framework for Early Detection of Sugarcane Pathogens via Image Processing on Embedded Hardware

Abstract

Precision Agriculture has become a revolutionary model on how crop productivity and sustainability can be improved. The crop organs that are most prone to diseases include red rot, red rust, mosaic virus and yellow leaf disease in sugar cane crops. This study suggests a synthesized artificial-intelligence system, which is a convolutional neural network (CNN) integrated with the MATLAB ResNet-50 architecture, deployed in the Simulink hardware, and edge processing in the Raspberry Pi. The system classifies sugarcane leaf disease real-time by image processing and deep learning. The suggested framework allows disease identification using the automated system, minimizes human control of monitoring, and aids in the implementation of early intervention measures. The experimental outcomes show a high classification accuracy and an efficient ability to operate in a real-time situation.

Introduction

Sugarcane is a vital global crop, serving as a primary source for sugar and bioenergy. However, its productivity is severely affected by plant diseases such as red rot, red rust, mosaic virus, and yellow leaf disease, which reduce yield and sucrose content, threatening farmers’ livelihoods.

Traditional disease detection relies on manual inspection, which is labor-intensive, subjective, and impractical for large plantations. This creates a need for automated, scalable, and accurate disease detection systems.

The integration of Artificial Intelligence (AI) and Internet of Things (IoT) technologies offers a precision agriculture solution. However, current systems face two major challenges:

1. Deep learning models are often designed for high-performance computing environments and are not optimized for low-cost edge devices like Raspberry Pi.

2. There is a lack of seamless frameworks that bridge model development (e.g., MATLAB) and real-time hardware deployment without extensive manual recoding.

Key Contributions

This paper proposes a unified AI-based framework with two main contributions:

1. End-to-End Deep Learning Pipeline

   • Uses ResNet-50 with transfer learning in MATLAB

   • Classifies five sugarcane leaf conditions: healthy, red rot, red rust, mosaic, and yellow leaf

   • Designed to generalize effectively despite small datasets

2. Edge Deployment via Simulink

   • Smooth integration of trained model into Raspberry Pi

   • Real-time, offline inference at the edge

   • Suitable for rural areas with limited connectivity

Related Work

1. Agricultural IoT and Precision Agriculture

While hyperspectral and robotic systems provide detailed plant analysis, they are expensive and complex. This framework instead uses low-cost RGB imaging, making it scalable and affordable for smallholder farmers.

2. Deep Learning Foundations

Deep neural networks, particularly ResNet-50, are effective in image recognition tasks. Residual connections solve the vanishing gradient problem and enable deeper architectures. Strict experimental design and baseline validation ensure reliable performance.

3. Resource-Efficient AI

Edge devices like Raspberry Pi have limited memory and processing power. Recent research emphasizes performance per resource unit (PePR) and optimized model design. This study applies transfer learning to balance classification accuracy with inference speed on constrained hardware.

Methodology

System Architecture

The proposed system is a closed-loop edge computing framework consisting of:

• Hardware Layer: Raspberry Pi 3 Model B+ with HD camera

• Software Layer: MATLAB for model training, Simulink for deployment

• Workflow: Image capture → Preprocessing → Deep learning classification → Result visualization/alert

Data Acquisition and Preprocessing

• Dataset includes five classes: Healthy, Red Rot, Red Rust, Mosaic Virus, Yellow Leaf

• Images resized to 224×224 pixels

• Pixel normalization applied

• Data augmentation (rotation, reflection, scaling) improves generalization and reduces overfitting

Deep Learning Model: ResNet-50

• 50-layer convolutional neural network

• Pretrained on ImageNet (transfer learning)

• Final layers modified for five-class classification

• Training configuration:

  • 70% training

  • 15% validation

  • 15% testing

  • Optimizer: Stochastic Gradient Descent with Momentum (SGDM)

Transfer learning reduces computational demand and improves performance with limited data.

Simulink-Based Hardware Deployment

A key innovation is seamless deployment using Simulink:

1. Trained MATLAB model loaded into Simulink

2. Simulink Support Package converts model into optimized C++ code

3. Code executed on Raspberry Pi’s ARM processor

4. Real-time loop:

   • Capture image

   • Resize and preprocess

   • Run inference

   • Output predicted disease

   • Trigger alert (e.g., LED or monitor display)

This enables offline, real-time disease detection directly in the field.

Conclusion

The paper includes a combined approach to precision agriculture that is centred on the sugarcane disease detection. Utilizing the advanced feature-extraction powers of ResNet-50 and the affordable Raspberry hardware and the Model-Based Design platform of MATLAB/Simulink, we can find the much-needed automated, real-time crop-monitoring. This strategy will reduce the inefficiencies of hand-inspections and will bypass the connectivity constraint of cloud-based solutions. Despite the existing issues related to environmental soundness and the interpretability of the models, the system is a step forward to scalable, resource-efficient, and affordable smart-farming technologies. This framework validation has a role to play in the greater aim of guaranteeing crop harvests and the economic sustainability of the sugarcane industry.

References

[1] Miller, T., Mikiciuk, G., Durlik, I., Mikiciuk, M., ?obodzi?ska, A., & ?nieg, M. (2025). The IoT and AI in Agriculture: The Time Is Now—A Systematic Review of Smart Sensing Technologies. Sensors, 25(12), 3583. doi:10.3390/s25123583. [2] Aldin, N. B. (2025). Real time detection of microbial leaf diseases using deep learning and edge computing on Raspberry Pi 4. Black Sea Journal of Engineering and Science, 8(4), 1185–1194. https://doi.org/10.34248/bsengineering.1704013. [3] Martí, Miquel, Maki, Atsuto, \"A multitask deep learning model for real-time deployment in embedded systems,\" Swedish Symposium on Deep Learning 2017, 2017. https://arxiv.org/pdf/1711.00146v1 [4] Shoaib, M., Khan, S. U., AbdelHameed, H., & Qahmash, A. (2026). Plant stress detection using multimodal imaging and machine learning: from leaf spectra to smartphone applications. Frontiers in Plant Science, 16, 1670593. doi:10.3389/fpls.2025.1670593. [5] Kawaguchi, Kenji, Kaelbling, Leslie Pack, Bengio, Yoshua, \"Generalization in Deep Learning,\" In Mathematical Aspects of Deep Learning. Cambridge University Press, 2022, 2017. doi:10.1017/9781009025096.003 [6] Berner, Julius, Grohs, Philipp, Kutyniok, Gitta, Petersen, Philipp, \"The Modern Mathematics of Deep Learning,\" Mathematical Aspects of Deep Learning, pp. 1-111. Cambridge University Press, 2022, 2021. doi:10.1017/9781009025096.002 [7] Mukhoti, Jishnu, Stenetorp, Pontus, Gal, Yarin, \"On the Importance of Strong Baselines in Bayesian Deep Learning,\" 2018. https://arxiv.org/pdf/1811.09385v2 [8] Ronen, Ronny, \"Why & When Deep Learning Works: Looking Inside Deep Learnings,\" 2017. https://arxiv.org/pdf/1705.03921v1 [9] He, K., et al. (2016). \"Deep residual learning for image recognition.\" Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 770-778. [10] Lu, Y., et al. (2017). \"Identification of rice diseases using deep convolutional neural networks.\" Neurocomputing, 267, 378-384. [11] Jiang, P., et al. (2019). \"CNN for image-based high-throughput plant phenotyping.\" Remote Sensing, 11(17), 2071.

Copyright

Copyright © 2026 Venkatesh S, Vijayakumar J, Mervin Paul Raj M. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.