PyriCane: A Large-Scale Benchmark Image Dataset for Smart Agriculture-Based Pyrilla Pest Object Detection, Infestation Identification, and Developmental Stage Classification in Sugarcane Crops

The proposed dataset is a large-scale annotated agricultural image collection designed to support hierarchical disease diagnosis and pest life-stage classification in sugarcane crops. The dataset contains 13,000 high-resolution field images acquired under real-world cultivation conditions, capturing variations in illumination, background complexity, leaf orientation, and infestation severity. These characteristics ensure robustness and practical applicability for computer vision and deep learning–based agricultural analysis. The dataset is structured to enable a two-stage hierarchical classification framework. In the first stage, images are categorized into Healthy and Diseased sugarcane classes, facilitating rapid crop health screening. Images identified as diseased are subsequently processed in the second stage, where the underlying cause of disease is determined by identifying the Pyrilla (Pyrilla perpusilla) infestation level, categorized into Egg, Nymph, and Adult stages. This hierarchical annotation strategy reflects real agricultural diagnostic workflows and reduces inter-class ambiguity commonly observed in flat multi-class classification schemes. All images are manually annotated by domain experts and undergo a quality-control process to ensure labeling accuracy and consistency. The dataset captures the complete life cycle of Pyrilla, enabling fine-grained pest severity assessment and supporting precision pest management strategies. The inclusion of multiple infestation stages allows researchers to investigate early, moderate, and severe disease progression scenarios. Owing to its scale and diversity, the dataset supports big data analytics and deep learning model development, including hierarchical classification, ensemble learning, and explainable AI techniques. The dataset is publicly available and intended to serve as a benchmark resource for research in plant disease detection, pest management, computer vision, and precision agriculture. 🔹 Optional Short Version (If Journal Has Word Limit) This study presents a large-scale annotated sugarcane image dataset comprising 13,000 field images, structured for hierarchical classification. The dataset supports two-stage analysis, where sugarcane plants are first classified as healthy or diseased, followed by identification of Pyrilla perpusilla infestation stages (egg, nymph, and adult). The dataset is designed to facilitate big data–driven deep learning research in agricultural disease diagnosis and pest management.本研究提出的数据集为大规模标注农业图像集，旨在支持甘蔗作物的分级病害诊断与害虫生活史阶段分类。该数据集包含13000幅高分辨率田间图像，采集自真实种植环境，涵盖光照差异、背景复杂度变化、叶片朝向不同以及侵染严重程度各异等多种田间场景。上述特性可使基于计算机视觉与深度学习的农业分析方法具备更强鲁棒性与实际应用价值。 该数据集采用两级分级分类框架结构。第一阶段将图像划分为健康甘蔗与染病甘蔗两类，以实现快速的作物健康筛查。经判定为染病的图像将进入第二阶段处理，通过识别扁蜡蝉（Pyrilla perpusilla）的侵染水平确定病害成因，具体分为卵期、若虫期与成虫期三个阶段。这种分级标注策略贴合真实农业诊断流程，可降低扁平式多分类方案中常见的类间歧义问题。 所有图像均由领域专家手动标注，并经过质量管控流程以确保标注的准确性与一致性。该数据集覆盖了扁蜡蝉的完整生活史，可实现精细化的害虫严重程度评估，支撑精准害虫管理策略。数据集包含多个侵染阶段，便于研究者探究病害在早期、中期与重度时期的发展进程。 凭借其规模与多样性，该数据集可支撑大数据分析与深度学习模型开发，包括分级分类、集成学习以及可解释AI（Explainable AI）技术。该数据集已公开可用，旨在作为植物病害检测、害虫管理、计算机视觉与精准农业领域研究的基准资源。 🔹 精简版（适配期刊字数限制） 本研究发布了一款大规模标注甘蔗图像数据集，包含13000幅田间图像，专为分级分类任务设计。该数据集支持两阶段分析流程：首先将甘蔗植株划分为健康与染病两类，随后识别扁蜡蝉（Pyrilla perpusilla）的侵染阶段（卵期、若虫期与成虫期）。本数据集旨在推动农业病害诊断与害虫管理领域的大数据驱动深度学习研究。