Analysis and Detection of Diseases in Leaf Images for Sugarcane using Efficient PARNET-52 Deep Learning Techniques

This research proposal aims to develop Deep learning algorithms for an IoT-based system to analysed and detect illnesses in sugarcane leaves. Using various deep-learning approaches, the system will detect and classify diseases by collecting leaf images using IoT-enabled sensors. The expected benefits of this strategy are increased gricultural
productivity, simplified early interventions and rapid access to reliable information for farmers. In addition to significantly improving precision farming methods, the proposed study could solve key challenges in agricultural disease prevention. To ensure maximum yields and minimize
financial losses in the agricultural industry, diseases in sugarcane crops must be identified promptly and accurately. Large-scale agricultural operations may be unable to use traditional disease detection techniques because they are often labour-intensive, time-consuming, and require
specialized knowledge. This study proposes an automated approach for applying deep learning methods and leaf image analysis to identify and categorize illnesses that affect sugarcane. A large dataset containing highresolution photos of sugarcane leaves with typical illnesses like red rot,
smut, leaf scald, and healthy leaves was gathered and annotated. Data augmentation methods like rotation, scaling, and flipping were applied to increase the dataset’s diversity and resilience. With its deep layers and effective feature extraction capabilities, the ResNet-50 architecture
is the Convolutional Neural Network’s (CNN) basis used in the suggested technique. After being trained and validated on the prepared dataset, the model achieved a classification accuracy of 97.2%. The model’s excellent performance demonstrates how effectively it can distinguish between different disease categories and healthy leaves, as shown by the low number of false positives and negatives. The study highlighted the specific leaf areas influencing classification decisions and provided a visual explanation
of the model’s predictions using gradient-weighted class activation mapping, or Grad-CAM. Gaining end-user trust and encouraging widespread adoption heavily depend on this interpretability. The results show that deep learning models provide fast, accurate, scalable diagnostic tools
and can completely transform scalable Sease management. This strategy will help farmers and other agricultural experts make informed decisions about crop management and treatment, resulting in increased crop yield and health. The model will be used in real-time applications like smartphone apps and drone-based monitoring systems, and future work will
concentrate on enlarging the dataset to encompass a greater range of diseases and environmental circumstances.