Red-Green-Blue (RGB) Image Classification Using Deep Learning To Predict Sugarcane Crop Age

Abstract

Traditional sugarcane growth monitoring methods are time-consuming and error-prone. This study investigated the use of deep learning to automate and enhance the accuracy of sugarcane growth stage classification. The study develops deep learning-based system that leverages high-resolution drone imagery for precise sugarcane age classification, thereby enabling accurate identification of the growth stages. High-resolution drone images were captured at various stages of sugarcane growth and were stitched together to form a comprehensive dataset. Segmentation of isolated areas of interest for analysis. The ResNet-50 deep learning model, enhanced with an additional fully connected layer, was used to classify the growth stages. The model was trained on cropped image sections, and its performance was compared to other deep learning architectures, such as GoogLeNet and VGG, to evaluate its accuracy. The ResNet-50 model outperformed other architectures, achieving 91% accuracy in classifying growth stages, demonstrating its effectiveness in agricultural image analysis and its potential to advance precision agriculture. This study is the first to apply deep learning to sugarcane age classification using high-resolution drone imagery, and it sets a new benchmark for agricultural image analysis. The dataset containing drone images from specific sugarcane fields may limit the model’s generalizability to different regions and environmental conditions.

Plant phenotyping relevance

ドローン画像からサトウキビの生育段階を推定する深層学習手法を開発し、複数モデルとの精度比較で検証しており、植物フェノタイピング手法が中心である。

Code and data availability

The paper's sugarcane drone image dataset (DJI Phantom 4 RGB images, age classes 2/4/6/9/11 months) is paper-specific, but no public repository or URL is given; the authors only offer to provide sample images on request. No code or model deposit is mentioned.

No evidence-backed public reproduction asset is currently recorded.