Utilizing UAV-based multispectral imagery and convolutional neural networks for brix value prediction

Abstract

Sugarcane is one of the major crops cultivated in tropical and subtropical regions worldwide. Assessing crop maturity is important for optimizing harvest timing and improving yield. Conventional sugarcane maturity evaluations use agronomic characteristics, past trends, and eye inspections, which are labor-intensive and not precise, particularly over large plantations. Some sugarcane varieties mature to complete ripeness earlier than their expected maturity age, rendering physical observation inefficient and unsuitable. To address this point, this experimental research study introduces a novel, cost-effective approach using Unmanned Aerial Vehicles (UAVs) equipped with multispectral sensors to estimate sugarcane maturity through remote sensing and deep learning techniques. The primary objective is to develop an efficient deep learning-based classification system for identifying mature sugarcane fields from multispectral images gathered using UAVs. Pelwatte Lanka Sugar Company (Pvt) Ltd geo-referenced yield data were used together with multispectral imagery of 3–12-month-old plant-crop sugarcane fields from intermediate and dry regions. Fields were classed as ‘matured’ (Brix > 10) or ‘immatured’ (Brix ≤ 10) based on mean Brix values. Red, Red Edge, Green, Near-Infrared (NIR), and spectral bands and vegetation indices NDVI and NDRE were investigated. 17,256 images with a resolution of 200×200 pixels were utilized (2,876 for each band/index). The dataset was split between training and validation sets. Modeling was done in two phases: (1) comparison of the feature extractor and (2) constructing a specific Convolutional Neural Network (CNN). The proposed CNN achieved a maximum accuracy of 93% on NIR images, whereas Red, Green, Red Edge, NDVI, and NDRE achieved 84%, 81%, 76%, 69%, and 59% accuracy, respectively. The results indicated that the model can classify sugarcane maturity with a high level of accuracy, thus improving precision agriculture methods.

Plant phenotyping relevance

UAVマルチスペクトル画像とCNNによりサトウキビの成熟状態（Brixに基づく）を推定する手法の開発・評価が中心であり、植物状態の取得・抽出方法に該当する。

Code and data availability

The paper describes a paper-specific UAV multispectral sugarcane dataset (17,256 labeled 200×200 tiles across five bands plus NDVI/NDRE) with Brix-based maturity labels and a custom CNN, but no public repository, deposit, or authors' URL for the imagery, labels, code, or trained model is provided. The data availability

No evidence-backed public reproduction asset is currently recorded.