Intelligent Sugarcane Plant Disease Detection using Deep Learning

Abstract

Abstract: Sugarcane is a vital crop that makes a substantial contribution to the agricultural economy globally, according to this study. Unfortunately, diseases that have a substantial effect on productivity and quality sometimes present a risk to its production. the system processes images of sugarcane leaves and stems. Combining visual processing and Manual inspections are used in most traditional disease detection techniques, which can be labour-intensive, time-consuming, and prone to human mistake. This paper provides a machine learning-based approach for sugarcane disease prediction that increases detection efficiency and accuracy by utilising Convolutional Neural Networks (CNNs) and environmental data. To visually recognise the signs of a disease, predictive modelling, the project aims to create an automated, real-time sickness diagnosis tool. Due to this tool’s ability to offer timely interventions, farmers will be able to lower crop losses and adopt sustainable agricultural practices. The proposed paradigm presents the agricultural community with a readily accessible and scalable alternative that could revolutionise crop health management. Data collection, pre-processing, augmentation, model training, and evaluation are some of the steps in the methodology. OpenCV, NumPy, and TensorFlow/Keras were used to handle image datasets, while Google Colab was used for training with GPU acceleration. The suggested model outperformed alternative CNN designs including VGG19, Xception, and ResNet50, with an accuracy of 91.94%. Gradio was used to create an intuitive user interface that allows users to upload leaf photos and receive immediate diagnostic feedback and confidence scores, enabling real-time illness identification.

Plant phenotyping relevance

サトウキビ葉・茎の画像から病徴を推定するCNN手法の開発・評価が研究の中心であり、植物病害状態の画像ベース表現型計測に該当する。

Code and data availability

The supplied blocks describe a DenseNet121-based sugarcane disease classifier, but contain no public dataset, image collection, code repository, or trained model availability statement with an authors' URL. All URLs in the blocks are reference citations, not paper-specific asset deposits.

No evidence-backed public reproduction asset is currently recorded.